WO2004013753A2

WO2004013753A2 - Data processing device and method for interactive multiple user access to large data volumes

Info

Publication number: WO2004013753A2
Application number: PCT/EP2003/008201
Authority: WO
Inventors: Peter Herold; Eckhard Kantz
Original assignee: Peter Herold; Eckhard Kantz
Priority date: 2002-07-26
Filing date: 2003-07-25
Publication date: 2004-02-12
Also published as: AU2003281782A1; WO2004013753A3; DE10234174A1

Abstract

The invention relates to a data processing device and to a method for interactive multiple user access to large data volumes within a given response time, comprising a plurality of computer nodes (40), each of which is equipped with its own working memory (44), and a load distribution and result synthesis device (20) for distributing at least one data processing process initiated by an interactive user query to a plurality of computer nodes (40) and for the synthesis of a full response to the interactive user query based on individual, partial responses supplied by the computer nodes (40), wherein the total data volume is initially broken down into disjunctive partial volumes, each of which is fully stored in the working memory (44) of a computer node (40) allocated thereto on a one-to-one basis, wherein the fraction of the data processing process initiated by the interactive user query that has been allocated to each computer node (40) is predetermined in such a way that it is fully completed within a length of time that is longer or shorter than a predetermined maximum length of time. Each computer node (40) generates its partial response on the basis of the partial volume of the total data volume allocated thereto and stored in its working memory (44) independently of all other partial volumes of the total data volume stored in the working memories (44) of the other computer nodes (40).

Description

Data processing device and method for interactive multi-user access to large amounts of data

The invention relates to a method and a data processing device for interactive multi-user access to large amounts of data. The invention further relates to a corresponding security analysis device and a security analysis method including a device and a method for automatically generating buy and sell signals for the management of securities in a securities account.

For numerous tasks in which means of information technology are used, it is important to include very large amounts of data from many GBytes (gigabytes) to several TBytes (terabytes) in a data processing process. In this context, for example, numerous approaches in the area of information search in structured data, also known as "data mining", are known.

In the prior art, however, significant restrictive technical limits are set with regard to the achievable values of data volume, response time and number of simultaneous user accesses when accessing large amounts of data.

In the analysis of the problem present in the prior art, problems and data structures from the area of the international securities market are used as examples. This in no way limits the applicability of the invention, but only serves the better Illustration. The solution can also be used for other applications where a predetermined small maximum response time is to be achieved while a large number of users simultaneously access extremely large amounts of data.

On the international securities market, well over a million securities are traded on a global scale. When assessing the opportunities and risks before buying securities, historical data on the course of the price are generally evaluated in order to draw conclusions about the likely development in the future. A distinction must be made in particular between the so-called "fundamental analysis", in which a systematic evaluation of data on the companies under consideration and their economic framework takes place, from the so-called "technical analysis", in which future price movements only use the share prices and their past sales figures be justified. For example, in the field of technical analysis, methods and parameters for so-called "chart analysis" are known

http://www.1stein.de/eckstein/indicators.htm Specifically, the following variables can be considered for use in securities analysis systems:

Moving Average (EMA Exponential Moving Average, trend following indicator): Calculates the average price of the last n trading days and thus draws a moving trend line. Depending on n, it follows the course of the course with a certain time delay.

• RSI (Relative Strength, cycle indicator): The simplest cycle indicator is momentum. It represents the momentum that the price movement currently has. The RSI is the further development of momentum with a standardized value range between 0 and 100. This makes individual securities comparable with one another. • Stochastics (overbought / oversold indicator): Stochastics is an extension of the RSI. It is assumed that the closing prices tend towards the high prices in bull market phases and towards the low prices in the downward trend. A trend reversal is imminent when the closing prices (with an upward trend) are no longer close to the

Maximum prices. Stochastics are very responsive and often lead the trend. It is therefore often smoothed or slowed down by a moving average of 3 days.

• MACD (Moving Average Convergence / Divergence, trend following indicator): With the MACD one tries to recognize the end of a prevailing trend in time. The MACD oscillates around its zero line. The stronger the trend, the further the MACD moves away from the zero line. The signals are best when the distance to the zero line is large. A signal is present when the MACD crosses the superimposed average line.

These methods are only given as examples here; numerous other methods of chart analysis are known. Such and other parameters can be calculated based on historical price data to prepare buying or selling decisions.

As soon as a large number of securities, including their historical data, are to be evaluated by a computer program, known information technology systems come up against limits due to the respective technological status in the field of processor speed, the speed of access to the working memory and the speed of the mass storage used, for example Hard drives are determined.

If a period of, for example, 10 years is to be taken into account for a retrospective analysis of the market activity on the securities market, the result is approximately 250 trading days per year and 1,000,000 different securities on the world market according to their respective security ID

250 trading days / year x 10 years x 1,000,000 securities = 2,500,000,000 "security days".

For each "security day", at least five data elements per security identifier must be saved for an evaluation using different methods:

• starting price of the relevant security on the relevant trading day,

• highest price of the relevant security on the relevant trading day,

• lowest price of the relevant security on the relevant trading day,

• the closing price of the relevant security on the relevant trading day, and • the number of shares traded (also referred to as "volume") of the relevant security on the relevant trading day.

It is now assumed that when using an optimized memory allocation for storing these five individual values, at least 8 bytes are required per security identifier and security day. With direct storage on a 32-bit architecture, 5 x 4 bytes = 20 bytes would arise. This results in a total amount of data of around 20 GB with optimized storage and around 50 GB with storage as a 32-bit value per security code and security day. Even the optimized amount of data far exceeds the maximum RAM expansion of conventional 32-bit architectures, which have their theoretical limit of the address space assuming a 32-bit addressing system at 4 GB. In practice, however, these architectures can provide even less for the application due to the memory requirements of the system software. When using a conventional single computer system, only the storage of all data on a hard disk with the corresponding storage capacity is possible. When carrying out a single evaluation process across all security days, access to each of the stored data elements must be realized. To do this, the total amount of data stored must be read from the hard disk. Even the fastest hard disk array currently available with an average linear reading speed of approximately 200 Mbytes / s will be the ideal case

20,000 MB / (200 MB / s) = 100 s

needed for reading. This shows very clearly that storing and reading the amount of data to be processed on a hard drive leads to runtimes that are two orders of magnitude higher than the response time of around one second required for interactive systems, even with a single evaluation process. This discrepancy increases further if several users access the system at the same time and want to carry out different evaluations over the entire amount of data.

Therefore, a value of one second is used as the target system response time. In the case of reading from a hard disk, this value can only be achieved due to the previously described high data volume if technological progress has increased the speed of read access to a hard disk or a hard disk array or to a suitable successive storage medium by a factor of around 100. In view of the extremely large amount of data, a system response time of one second can therefore only be achieved in the foreseeable future by keeping all the data in memory. But ^'also the current technological state has its limits at the maximum memory expansion as well as the processor clock frequency.

The minimum memory expansion of 20 GB required to solve the problem outlined above goes beyond the limits of 32-sit architectures and is therefore only practical in practice as a single computer system when using a 64-bit Architecture feasible.

In this case, however, the processor clock frequency and in particular the maximum access speed to the main memory become a limiting factor. If the amount of data to be stored is changed from 2,500,000,000 security days to the current processor speed of around 2,500,000,000 Hz, the result is that on average only one processor clock is available per security day within one second. This size is therefore not a sufficient basis for the execution of relevant processing operations. Thus, the use of 64-bit architectures would not solve the problem with the current state of the art.

For the further considerations, a value of 20 processor clock cycles based on practical experience is used as the minimum number of processor clock cycles for carrying out a minimal processing of the data of a single security day. The minimum processing to be performed with this number of cycles is the reading out of an individual characteristic value from a security day, the comparison with a predetermined comparison value and the derivation of a decision therefrom. More complex processing usually requires significantly more processor cycles. The number of processor cycles usually depends very much on the internal architecture of the processor, in particular on the parallel processing of instructions. A common approach to overcoming processor capacity bottlenecks is to use vector computing architectures that are capable of processing multiple elementary operations in parallel. Parallel processing, however, has theoretical limits, particularly when executing multiplication and division instructions, which are predetermined by the internal architecture of the arithmetic processing unit of the processor and which cannot be influenced by any programming optimizations in the application program. For the selected minimum value of 20 processor cycles per security day, with a required response time of one second, a simple calculation results in a required processor clock frequency of

20 cycles / security day * 2,500,000,000 security days / s

= 50,000,000,000 cycles / s

= 50 GHz (clock frequency).

Assuming that Mopre's law, which postulates a doubling of the processor clock frequency after every one and a half years, will remain valid in the next few years, it will take a period of at least 6 years until technological progress increases the corresponding Processor clock frequency produces.

The calculation of the performance limits in relation to a specific technological status is described below. Two relations are to be evaluated, each describing an aspect of the limitation due to the technological state:

1) Limitation by processor speed

Z * D * B <T * F * P (1)

It means:

Z number of clock cycles for the determining algorithm of the functionality under consideration;

D number of data units to be processed for a user action;

B number of users who simultaneously access the functionality under consideration; T Maximum response time of the system when calling up the functionality under consideration;

F processor clock frequency (number of processor cycles per second); and

P Number of processors used in parallel.

As soon as five of the named parameters are specified, the sixth parameter can no longer be freely selected, but is then subject to the limitation by the relation mentioned.

As an example, the previously assumed values are used to calculate the response time for a single user when using only one processor: For

Z = 20 cycles / (security ^* user),

D = 2,500,000,000 securities,

B = 1 user,

F = 2,500,000,000 cycles / (s * processor),

P = 1 processor

surrendered:

T> Z * D * B / (F * P) ~

T> 20 cycles / (security * user) * 2,500,000,000 securities * 1 user / (2,500,000,000 cycles / (s * processor) ^* 1 processor)

T> 20 s Result: The response time will be at least 20 seconds.

2) Limitation by the access speed to the working memory

S ^* D * B <T * V * P (2)

It means:

S memory requirement for a data unit to provide the functionality under consideration;

D Number of data units to be processed for a user action.

B number of users who simultaneously access the functionality under consideration;

T Maximum response time of the system when calling up the functionality under consideration;

V Speed of the main memory when reading data; such as

P Number of processors used in parallel.

As an example, the previously assumed values are used to calculate the response time for a single user using very fast working memory with an access speed of 2 GByte / s with only one processor and optimized storage layout of the data:

For

S = 8 bytes / (security day * user), D = 2,500,000,000 securities, B = 1 user,

V = 2,000,000,000 bytes / (s * processor), P = 1 processor

surrendered:

T> S * D * B / (V * P)

T> 8 bytes / (security day * user) * 2,500,000,000 security days ^* 1 user / (2,000,000,000 bytes / (s * processor) * 1 processor)

• T> 10s

Result: The response time will be at least 10 seconds.

It becomes clear that there are limitations for the example parameters used from both relationships, which are of the same order of magnitude. However, this does not necessarily have to be the case in other applications. As soon as an application has a significantly higher complexity of the algorithms, there will be a limitation due to the processor speed.

If, on the other hand, the storage space requirement for a smallest data unit that repeats itself billions of times increases, then the limitation will be dominated by the access speed to the working memory.

In the future there will be ever increasing demands on a process and systems for the rapid processing of large data volumes. On the one hand, there will be ever larger amounts of data that should be accessed within a period of time that is acceptable for interactive operation. On the other hand, an increase in the number of users is expected, which is more Operate systems. Ultimately, new knowledge will increase the demands placed on the algorithms used, which results in an increased need for processor cycles per smallest data unit, in the example considered, therefore, per security day.

In the following, a factor of 1,000 (one thousand) is assumed as a hypothetical technological leap for the technical parameters processor speed, access speed to the main memory and access speed to the external memory.

3) Processor speed a thousand times faster than at the beginning of 2002

For the parameters already used, it is determined how many users can access the system at the same time, assuming a maximum response time of 0.2 seconds (without delay from the user's point of view), if only one processor with 1000 times the speed of today's systems is used:

For

Z = 20 cycles / (security * user),

D = 2,500,000,000 securities,

T = 0.2 s,

F = (1,000 * 2,500,000,000) cycles / (s * processor),

P = 1 processor. -

surrendered:

B <T * F * P / (Z * D) (3)

B <0.2 s * (1,000 * 2,500,000,000) cycles / (s * processor) ^* 1 processor / (20 cycles / (security * user) * 2,500,000,000 securities) B <10 users

Result: A maximum of 10 users with a response time of 0.2s can access the system simultaneously, even if the processor speed would hypothetically increase a thousandfold.

4) Memory access speed a thousand times higher than in 2002

For the values used in calculation example 2, the response time is now calculated if the access time to the working memory would improve by a technological leap by a factor of 1,000 (one thousand) to (1,000 * 2) GByte / s and 100 users simultaneously on the Access system.

For

S = 8 bytes / (security day * user),

D - 2,500,000,000 security days,

B = 100 users,

V = (1,000 ^* 2) GByte / s,

P = 1 processor

surrendered:

T> S ^* D- * B / (V * P) (4)

T> 8 vByte / (security day * user) * 2,500,000,000 security days * Tö0 user / (1,000 * 2,000,000,000 bytes / (s * processor) ^* 1 processor)

T> 1 s Result: The response time for 100 users will be at least one second, even if the access speed to the RAM would hypothetically enlarge to a thousandfold.

5) Access speed to the external memory 1000 times higher than in 2002

5 The response time is now calculated for the values already used in calculation example 2, as in calculation example 4, if the access time to the external memory would improve by a technological jump by a factor of 1,000 (one thousand) to (1,000 * 200) Mbytes / s and continue to access 100 users at the same time.

o For

S = 8 bytes / (security day * user),

D = 2,500,000,000 security days,

B = 100 users,

V = (1,000 * 200) MByte / s, 5 P = 1 processor

surrendered:

T> S * D * B / (V * P)

T> 8 bytes / (security day * user) ^* 2,500,000,000 security days * 100 users / (1,000 * 200,000,000 bytes / (s * 0 processor) * 1 processor)

T> 10s

\

Result: The response time for access by 100 users will be at least 10s, even if the access speed to the external memory 5 would hypothetically increase 1000 times.

The results of the calculation examples 4) and 5) show that even after a big leap in technology, the access speed increases Storage media by a factor of 1,000, keeping the extremely large amount of data in the external memory in terms of throughput does not match the speed when storing the amount of data in the working memory, as long as the speed for accessing working memory and external memory increases to the same extent.

Furthermore, the calculation example 3) makes it clear that every technological advance is divided into several dimensions, such as the number of users and response time, when it is used in practice, and therefore only a fraction of the factor which makes up the respective progress is accounted for in each dimension.

A division of the computing work to be carried out over several computers working in parallel will therefore also make sense in the case of technological progress with a factor of 1,000 and beyond.

A multiprocessor system is known from US-A-5,734,829, which increases the speed for a processing algorithm of seismic data to a multiple of the value of individual processors by working in parallel with several processors. The group of processing processors is controlled by a central module, which assigns the processing processors the appropriate data records for processing and gets the result back and forwards it.

In contrast to the present problem, there is no criterion for processing an extremely large amount of data in a given short time in the situation described in the prior art document; Instead, the intention is rather a general increase in performance compared to a single processor system.

The system concept "BeoWulf from http://www.beowulf.org/ and existing implementations of this concept continue to be regarded as state of the art

System concept BeoWuif is understood as the interconnection of two or more computers via a local network to form a cluster. The goal of Clusters also consist in achieving a significant increase in performance compared to a single computer system.

In such a BeoWulf system there is always a control program which appropriately divides a processing request into subtasks, which are then sent to the computers in the BeoWulf system for parallel processing. Furthermore, the processing results of the individual computers are returned to the control program and are combined to form a common answer. The composite common response is then sent to the sender of the original request, completing the processing of the request.

If the task at hand in the sense of a BeoWulf system were to be solved by interconnecting ^' several computers, then when estimating the number of computers required it can be assumed that the data is kept, for example, in a suitable, very fast database, but nevertheless due to the extremely large amount of data can be read from the hard disk during evaluation. As previously determined, the fastest hard disk array currently available has a response time of 100 seconds. In order to achieve a response time of one second, a total of 100 such hard disk arrays would have to be interconnected, while maintaining all the assumptions previously made, in order to achieve the required response time.

The requirements for the number of hard disk arrays are reduced if the accelerating effect of the cache is taken into account. The problem that arises here, however, is the provision of sufficient cache to effectively keep all the data in the working memory and, furthermore, in particular the runtime overhead and its imponderability due to the cache management algorithms, which are generally not known. If, for example, the cache is viewed by database servers, it has to make a compromise between the requirements of the most varied access types and is therefore usually an overhead for each individual case each have optimal access procedures to the data. A quantitative assessment of the overhead is only possible in the specific case depending on the manufacturer of the database and usually also on the version of the software used. This makes it difficult to achieve a defined short response time with extremely large amounts of data.

The object of the invention is to provide a method and a data processing device for interactive multi-user access to large amounts of data, which in particular do not have the disadvantages specified above or only have them to a greatly reduced extent. In particular, it is also an object of the invention to provide a method and a data processing device for interactive multi-user access to large amounts of data, which scale excellently even with given maximum response times under extremely large amounts of data and numerous users. Finally, it is also an object of the invention to provide a corresponding security analysis device and a security analysis method.

This object is achieved according to the invention by a data processing device with the features specified in independent claim 1 and also by the method with the features specified in independent claim 4. Furthermore, this object is achieved by a securities analysis device with the features specified in claim 5 and by a securities analysis method with the features specified in claim 10. The subclaims contain further developments of the subject matter of those independent claims to which they are in each case related. The inventive method shows how predetermined values for the three sizes can be achieved through the use of computers working in parallel by means of sufficient working memory, and a data processing device according to the invention for carrying out the method is specified. In the solution according to the invention, in contrast to conventional cache algorithms, the total amount of data is divided over the Main memory of several computers and full control over the majority of the processor cycles required for processing achieve a defined short response time of the computer cluster even with extremely large amounts of data. The distribution of the total amount of data to be selected in the specific case

5 computers in a cluster must be such that the two relations (1) and (2) explained above are fulfilled simultaneously for each individual computer. This limits the applicability of the method to such data, where such a division is possible due to the internal structure of the extremely large amount of data, as is the case, for example, with trading data for o securities.

Furthermore, a successful application of the method requires that the number of processor cycles for the determining algorithm is determined. This is done, for example, by programming the system time before the algorithm starts for a single data unit, and determining the system time after the algorithm has ended for this one data unit. The number of processor cycles required is calculated from the knowledge of the processor speed and the previously determined time required for processing the algorithm.

Because of the limited resolution of the system time of, for example, 0 milliseconds, it is necessary to run the algorithm very often in succession, for example 100,000,000 times, in order to obtain usable time differences. The number of processor cycles searched is then calculated as follows:

Z = F ^* (T2 - T1) / N

Z number of cycles for the implemented algorithm; 5 F processor clock frequency = number of

Processor cycles per second;

T2 end time of measurement;

T1 start time of measurement; N Number of repetitions of the algorithm.

If, for example, an implemented algorithm has a time difference of

T2 - T1 = 0.8 sec. at a processor clock frequency of 2,500,000,000 cycles / s and a number of N = 100,000,000 repetitions, this results in a number of cycles of:

Z = 2,500,000,000 cycles / s * 0.8 sec. / 100,000,000 = 20 cycles.

Result: 20 processor cycles are required to execute the implemented algorithm with a single data unit. As a result of the application of the present method, in contrast to cache algorithms in databases, a defined runtime is also achieved for an extremely large amount of data when using algorithms with a known need for processor cycles. The unpredictability of the database cache is eliminated and instead the system always offers a defined response time. The invention is illustrated below by way of example using one or more exemplary embodiments in a "message passing" architecture with reference to the drawing:

1 shows a schematic hardware and software component diagram of an embodiment of a data processing device according to the invention based on the "message passing" principle.

Fig. 2 shows schematically the hardware components of the embodiment shown in Fig. 1.

3 shows a schematic block diagram of an individual computer (computer node) from the embodiment shown in FIG.

FIG. 4 shows a flowchart of the algorithm for load sharing and result synthesis from the embodiment shown in FIG. 1. 5 shows in the upper part a diagram with typical typical buy and sell signals for short and long positions of a security shown on a course of the price with the associated earnings diagram in the lower part.

Fig. 6 shows the course of optimization and short and medium-term test phases to determine the optimal parameters of a trading system.

Fig. 7 shows the assignment of several trading systems and trading variants to individual, different securities in a securities account.

Fig. 8 shows a linear trading system that corresponds to the prior art.

Fig. 9 illustrates the complex evaluation of a large number of securities taking into account a large number of trading systems and taking market conditions into account.

For the application of the present method in a technical system, a computer cluster shown schematically in FIG. 1 makes its functionality available to other systems via an external network 10. These can be, for example, inquiries from people (not shown) who use a browser (not shown) to call up information on the Internet via a graphical user interface. Likewise, these can be other computers or computer systems that can be connected to the system described here via the Internet and that queries can be carried out via a machine interface. In addition to the Internet, the external network can also be any other data network that connects other computers to the system described. A number 1 ... n individual computers 40 form a group of computers which form the processing units of the system. In a preferred embodiment, a load distributor 20 routes requests from an external computer network 10 to the individual computers 40 using a suitable distribution strategy further, so that the load is divided among the individual computers 40. The load distributor 20 communicates with the individual computers 40 via an internal network 30. Instead of a load distributor with optimization functionality, a simpler load sharing device can otherwise also be used.

The only requirement for the load distributor 20 is that the incoming requests are suitably distributed among the individual computers 40, so that on average a uniform load on the individual computers 40 occurs. It is further assumed that this task is carried out by a commercially available device of this type, the algorithm running there not being considered in more detail here.

Load distributors are offered by different providers with different properties. For the construction of a server farm according to the present description, the "Catalyst 4840G" load distributor was selected as an example, which allows the incoming requests from the external network to be distributed to a maximum of 40 servers. The "round robin" method is used as the load distribution strategy, which is a uniform method Load on all connected servers. A description of this load balancer can be viewed at the following Internet address:

http://www.cisco.com/univercd/cc/td/doc/pcat/ca4840g.htm

In the example configuration given, the load distributor ensures an even load on all servers, which is necessary for the continuous operation of a large number of 600 requests per second and more from the external network. However, this has no direct connection to the defined short response time, which the system also adheres to for an extremely large amount of data.

The short response time is achieved in particular by the fact that each computer in the cluster is allocated only a part of the extremely large amount of data for processing and that it keeps this amount of data in its main memory to enable the generation of a quick response. A distinction must therefore be made between two extreme situations. In the case of a complex request which requires the complete amount of data to be viewed, the algorithm according to the invention in accordance with FIG. 4 is used. By dividing the required work across all connected computers in the cluster, the intended short response time to the complex request is achieved.

In the other extreme, less complex requests are made, for example, for a single course, but with a large number of 600 requests per second and more. In this case, the conventional load balancer ensures that requests are evenly distributed to all computers involved. The system is able to respond to a single complex request that requires all data to be viewed within a predetermined time, for example one second, and to generate the response. Here, all computers are fully used up to their performance limit. For the reason mentioned, it will generally not be possible to combine both extreme cases with each other. In practice, however, a mixture of requirements can be assumed, where a few complex requests occur within a large number of less complex requests. The frequency of the complex requests must be significantly less than the response time that was configured to answer a single complex request. 1 shows the hardware and software modules relevant for the overall function within the individual computers 40. A server program 41 accepts requests that have been made to the system from the external network 10 via the load distributor 20. Each request is forwarded to an intermediary program 42 on the same computer. The broker program 42 sends each received request to a request processing program 43 on the same computer as well as via the internal network 30 to the broker programs 42 on all other individual computers 40. The request processing program 43 accepts each request from the intermediary program 42 on the same single computer 40 and generates a corresponding answer. To generate the response, all data are stored in a memory 44 or, depending on the request, only parts of it used.

The mediator program 42 accepts the answer to a query from the query response program 43 on the same individual computer and combines this with the answers from the mediator programs 42 on all other individual computers 40 to form a combined answer. The combined response is forwarded by the intermediary program 42 to the server program 41 on the same individual computer, which sends the received response to the requesting system in the external network 10.

By using the distributed individual computers 40, several positive effects are achieved. On the one hand, the entire extremely large amount of data that is to be made available interactively for multi-user access can be distributed over a plurality of individual computers 40 in such a way that each individual computer 40 receives a quantity of data that can be kept within the memory 44 at the same time and that is included in the processing allows the proposed algorithms a response time less than or equal to the specified response time. Furthermore, the tasks of the server program and the intermediary program are divided among all available individual computers, which enables an optimal load distribution in the overall system.

The maximum amount of data that can be placed on a single computer 40 depends, as previously described in detail, on the high processor speed, speed in accessing the working memory, intended response time and the number of cycles required to execute the intended algorithm for a single data unit.

Before a single computer can answer queries, the amount of data relating to this single computer must be loaded into the memory 44. This happens from the hard disk 45 when the individual computer is started up from the idle state to the standby state. Due to the limited hard disk speed, this process takes several orders of magnitude longer than the later processing of the entire amount of data in the memory 44. The data must therefore be loaded in good time before the first request is processed. When the system is in operation, newly arriving data, such as, for example, the closing prices on the current trading day, must be transferred both to the memory 44 and to the hard disk 45. If the permissible amount of data for the individual computer 40 is exceeded, older data must be deleted from the memory at the same time.

In one embodiment used in the field of securities trading, a special data compression technique described below can be used. This preferred optimized memory allocation takes advantage of the fact that the values for • the initial course (Open),

• the high (high),

• the lowest price (Low), and / or for

• The closing price (close) is very close to each other on most trading days. Hence the differences

• open-low,

• Close-Low and

• High-Close saved in one byte each. A compression scheme is given here as an example among several possibilities, which reduces the storage space at the expense of resolution (accuracy). However, this only occurs if the price movement is dynamic or if the trading volume is very high. In the vast majority of cases encountered on the securities market, this procedure is lossless. However, the individual values must meet the following conditions:

• The closing price must not be greater than 65,535,000, and »the volume must not be greater than 655,350,000

Because the numbers representing the market values are more relevant Currency can take on different orders of magnitude, a floating point representation is preferred, in which a numerical value is represented by a mantissa value and an exponent value. In a preferred embodiment, the absolute value of the closing price is stored in addition to the course differences mentioned above (open-low, close-low and high-close). This means that the differential values, which are stored in a particularly economical memory, can be converted back into absolute market values:

Open = Close - (Close - Low) + (Open - Low)

Low = Close - (Close - Low) High = Close + (High - Close)

Under the specified conditions, the individual values are stored in the memory as follows in the compression process described here:

• 1 byte: exponent

• Bit 7-5: Common exponent to base 2 for the 3 - difference values, (Open-Low, Close-Low and High-Close;

Range of values of the exponent 0..7)

• Bit 4-2: Exponent to base 10 for all 4 price values, (i.e. the above-mentioned difference values and closing price value; value range of the exponent + 3..0 ..- 4) • Bit 1, 0: Exponent to base 10 for the volume,

(Range of values of exponent 1..4)

1 byte: difference open - low (value range of mantissa 0..255)

1 byte: difference close - low (value range of mantissa 0..255)

1 byte: difference high - close (value range of mantissa 0..255) »2 bytes: close (value range of mantissa 0..65535)

2 bytes: volume (value range of the mantissa 0..65535) Example: AOL courses on February 28, 2002 • Open = 24.22

• High = 25.50

• Low = 23.70

• Close = 24.80

• Volume = 28,516,800

The following integer values to be stored result for the example values used:

Table 1: Examples of compression methods (compression step)

e values are stored in the memory as follows:

Byte 1, bit 7-5: 0 (=> 0, exponent to base 2 for the difference values) byte, bit 4-2: 5 (=> -2, exponent to base 10 for all price values) byte 1 ^ bit 1, 0: 2 (=> 3, exponent to base 10 for the volume)

Byte 2: 52 (open-low)

Byte 3: 110 (close-low)

Byte 4: 70 (high-close)

Byte 5,6: 2480 (Close) Bytes 7,8: 28517 (volume)

To check the accuracy of the values after decompression, the course values and the volume are subsequently calculated from the stored values. It becomes clear that the market values could be reconstructed without errors, whereas the volume has a small error, which is negligible for the intended application.

Table 2: Exemplary compression methods (decompression step)

FIG. 2 shows an example of the application of the system described above to enable access to an extremely large amount of data with a defined short response time for users who access the system via the Internet 200. In addition to a load distributor 210, a number N of individual computers 220 are used, so that the desired short response time is based on the previous calculation methods.

3 shows the components of a single computer that are necessary for achieving the intended solution. Within the single computer 300, a processor 340 works together with the working memory 305, a hard disk 345 and a network card 330 via the system bus 335. System programs 310 are located within working memory 305 at runtime. Application programs 315 and a data store 320, which takes up the majority of the main memory.

An example computer configuration of an embodiment according to the invention is given with reference to FIG. 3. The following components are used:

300 - Computer, mainboard: Chaintech, 9SJD Intel P4

305 RAM: 3 x 1GB DDR PC2100, 266 MHz RAM

325 - Local network connection: Ethernet, 10/100 Mbit

330 - Network card: 3C905C, Fast EtherLin XL PCI 340 - Processor: Intel P4 2.2 GHz, socket 478

345 - Hard drive: IBM IC35L120AWA 120GB

The algorithm for processing requests is described in detail below with reference to FIG. 4.

In the following, a message is understood to mean the arrival of an input data packet at the intermediary 42. This can be both an inquiry message from the server component 41 and an inquiry or response message from another computer 40 of the cluster.

After start 400, each of the messages listed above is received by input module 405 and forwarded to decision block 410. There it is decided whether it is an inquiry or an answer. In the event of an inquiry, it is forwarded to decision block 415. There it is determined whether the request concerns the cluster as a whole or the computer itself. If the request has been made to the cluster, block 420 forwards it to any other computer in the cluster. In addition, the same request is given by block 425 as a local request to decision module 430, which is also addressed by decision module 415 when a request arrives at this computer. Decision block 430 determines whether the request can in principle be processed further based on the locally stored data or not. If none for editing data suitable for the request is then stored, block 440 continues where an empty response is generated. This ensures that a response is always sent back to the request divider.

In block 435, if suitable data is available in the local memory, the query is processed in accordance with the specific content of the query. An algorithm for processing data units is used, the need for processor clocks has preferably been analyzed in detail beforehand.

After _. Generating a response to the request in block 435 or, if appropriate, an empty answer in block 440 continues with decision block 445. There it is determined whether the request has arrived from another computer or whether it was generated locally. If the request was sent by another computer, the response is sent back to the requesting computer by block 460 and the algorithm in end block 465 is thus ended.

On the other hand, in the case of a local request, the response to that request is forwarded to block 450 as part of the overall response. Block 450 assembles all the answers arriving from other computers via decision block 410, including the answer generated locally in block 435 or 440, into a common answer. This also includes caching partial responses at least until all responses have arrived from all computers in the cluster and these can be combined to form a complete response.

Decision block 455 checks for the aforementioned reason whether the current response is already a complete response after the addition of the last partial response that has arrived. If this is the case, block 460 sends this to the sender of the request and the algorithm ends with end block 465. Otherwise, it is a fragment of the entire response, which is still stored locally, by further incoming responses from the other computers in the cluster, or also by local request processing later by a new run of the algorithm to be added. The current instance of the algorithm is therefore ended with end block 465.

As can be seen from the algorithm described, the sending of the composite response depends on the arrival of all partial responses. By introducing a suitable timeout time, it will therefore be necessary to check whether the computer cluster as a whole is functional. When the timeout period is reached, the functionality is no longer present and appropriate error handling takes place in an automated manner or manual intervention is carried out, if necessary.

Application example for a depot administration

As an example of an application of the security analysis system based on the previously described method and the data processing device for interactive multi-user access to large amounts of data, a method and a device for the automatic generation of buy and sell signals for securities in a securities account are described below, the aim of which consists in generating a positive return.

The typical task in managing a custody account is to convert the amount of capital available on a certain day into security purchases. In the system described here, formulas are used that are able to generate buy and sell signals. These formulas are called trading systems. They are created by adding logical conditions for buy and sell signals to indicators and chart histories, for example exceeding lower or upper limit values. Figure 5 shows an example of the work of a trading system. Based on the past and current price data of a security, this generates signals for opening long or short positions on certain days. A long position is used when securities are bought and held in the securities account for a certain period of time. At a later time, the long positions will be sold again with the aim of taking more than when selling as was spent on the purchase. The difference between the sale and purchase amount is the income from this position, whereby costs that are related to the purchase and sale processes must also be taken into account. The yield can be both positive and negative. In contrast to a long position, a short position sells first and repurchases later. In order for this to be feasible, the number of securities to be sold must first be borrowed from someone else, who usually gets the securities back when the position is closed. There are additional rental fees for borrowing. A positive return can only be achieved with a short position if the price of the security has fallen sufficiently in the meantime. Short positions are therefore used to bet on falling prices.

The generation of buy and sell signals using formulas can be done directly on the basis of the price movements of the security (chart development) or the trends of indicators, such as a moving average, can be used. A simple trading system can consist of two moving averages, which are calculated over different periods of time. A buy signal is generated when the shorter moving average crosses the line of the longer moving average from bottom to top. In this case, an upward trend of the security is displayed. In the opposite case, a sell signal is generated analogously.

The return is determined when a position is closed, ie when a long position is sold or when securities are bought to close a short position. Figure 5 shows an example of how a positive return adds up when a long and a short position are subsequently closed. In the case of negative yields, the yield curve drops accordingly, as can be seen in part in Figure 6.

A trading system can usually be adjusted to the market conditions using parameters. In the case of a moving average, the number of days averaged over is such a parameter. With the change The yield curve of a trading system also changes from parameters because the buy and sell signals shift and the securities are therefore bought or sold at different prices.

Optimizing a trading system means determining those parameters that bring the yield curve into the desired shape and height. For this purpose, a brute force approach, for example, calculates all the possibilities of parameter combinations and determines such a parameter set that delivers the highest value for a target function, for example the highest yield. In practical use, optimizing for the highest yield does not necessarily lead to a trading system that will continue to generate good returns in the future. It is often more successful to optimize a trading system in such a way that as little as possible during the optimization period and then only a small negative yield. The evaluation of an optimized trading system as a basis for decision-making for use in depot management is carried out using two test periods, a longer test period ti of several months and a shorter test period t ₂ of a few weeks. The longer test period ti follows the optimization period to and is used to determine similarities in the behavior of all trading systems that are used for a given security. The longer test period ti extends to the current day. The shorter test period t _{2 also} extends to the current day, but only covers a short past period. Examples of the length of the periods are 5 years for the optimization period to, 6 months for the longer test period i and 2 weeks for the shorter test period t ₂ .

The complete portfolio management works with several trading systems per security and several variants per trading system as shown in Figure 7. This starts with a simulation of real market events, where traders usually make their decisions to buy or sell securities with the support of securities analysis systems. Furthermore, it is shown how with the help of statistical Methods the most promising trading systems are currently being determined and how decisions on the use of capital in the portfolio are derived from them.

Before doing so, however, it must be explained how several variants of a trading system can be created for a given security. When using the brute force method, this is obviously not the case, because only one variant of a trading system with a given target function is obtained here. With the brute force method, there are several variants whenever the target function is varied accordingly, for example once with maximum yield and in another run with minimal interim loss (also called drawdown). However, this method will generally take a very long time, for example several hours or days or weeks, for each individual optimization run, depending on the number and value ranges of the parameters, or it is possible for large parameter numbers by a theoretical runtime of Example several thousand years outside of the current technical possibilities.

In the present case, therefore, the method used to find local maxima known in mathematical optimization is used, which, by evaluating the gradients of the target function, already delivers a result for each individual parameter in a few hundred steps. After the first local maximum has been found, the search process is continued in such a way that a further local maximum of the target function is found, but this is not identical to the previous results. By means of a computer configuration as in examples 1) and 2), a time of on average 5 seconds is required for optimizing a trading system using the local maximum method. In the case of a portfolio with, for example, 6000 securities and the use of 150 trading systems in 10 variants each for rising trend (bull market), falling trend (bear market) and for sideways markets, this results in the following time estimate: 6000 securities x 150 trading systems x 3 market situations x 10 variants x 5 sec.

= 135,000,000 seconds

= 37,500 hours = 1562.5 days

A total of 27 million different trading systems are generated, which form the basis for statistical evaluations based on them. The reduction of the time necessary for the optimization to a practical value is carried out by using several computers according to the method described above. Using 50 computers, for example, results in a time requirement of

1562.5 computing days / 50 computers = approx. 31 days.

This means that in this case the generation of trading systems must begin at least one month before the start of trading of values from the securities account, so that a sufficient number of trading systems are available as the basis for statistical evaluations at the start of trading.

The most important criterion for considering the signals of a trading system when using capital in the custody account is that no negative return occurred in the second test period t ₂ , i.e. in the immediate past. All trading systems with a currently negative return in the test period t ₂ are therefore the first to be eliminated from further consideration for the current trading day. There remain trading systems with a positive return and those trading systems which did not generate any signals in the relatively short test period t ₂ .

Next, an average yield curve for the test period ti is calculated by averaging the daily values of all trading systems with a yield greater than zero in the test period ti. The average yield curve thus represents the behavior in the longer test period ti of all trading systems that are successful under current market conditions in the shorter test period t ₂ .

The selection of trading systems is then carried out according to how well a trading system matches the behavior of the average trading system of all currently successful trading systems. For this purpose, the well-known mathematical method of correlation is used, in that it is applied to the yield curve of a trading system in the test period ti and to the averaged yield curve of all currently successful trading systems. If the yield curve of a trading system agrees well with the averaged yield curve, a correlation value in the vicinity of 1.0 is calculated, whereby 1.0 cannot be exceeded. If, on the other hand, the trading system has a yield curve with the opposite course, then a value of -1, 0 is used for the _. Correlation received, whereby the value cannot fall below -1, 0. A value of 0 indicates that there is no obvious correlation between a trading system's yield curve and the average yield curve.

After a correlation value between -1 and 1 has been assigned to all trading systems, the final selection of the trading systems to be used on the current trading day is made by applying a threshold value, for example the correlation must be greater than 0.85. This procedure leads to a reduction of the previously determined group of trading systems with zero value or positive earnings in the test period t ₂ to trading systems which behave similarly to the currently most successful trading systems.

The use of this statistical method leads to the inclusion of trading systems that currently have good chances of success, but have not generated any signals in the relatively short test period t ₂ . Because of their similarity to the currently successful trading systems, these trading systems potentially have a high probability that a profitable transaction is imminent. All trading systems determined in this way then determine how many of them have generated a buy signal on the current day. The number The trading systems determined represent the weight that the respective security receives when the amount of capital to be invested is invested.

As an example, assume that five securities are being analyzed. The number of buy signals determined using the method described above should be distributed as follows:

1st security: 30

2nd security: 0

3.Security: 12

4th security: 18 5th security: 0

A capital to be invested available to an investor of, for example, EUR 20,000 is then divided as follows:

1st security: 30/60 * 20000EUR = 10000 EUR

2nd security: 0 3rd security: 12/60 * 20,000 EUR = 4,000 EUR

4th security: 18/60 * 20,000 EUR = 6,000 EUR

5. Security: 0

The capital invested in a security is allocated to the individual trading systems to manage open positions. In the case of the example above, each open position receives an amount of 20000/60 = EUR 333.33 or more precisely securities with this daily value. It is rounded to whole pieces. When a trading signal is generated by the respective trading system and only then is the number of securities that has just been determined sold again and the proceeds obtained are made available for a new investment, for example.

When allocating an existing amount of capital to a larger number of securities in a securities account, it must be taken into account that buy signals from different securities over a period of, for example, some Days will be distributed and that the total capital can therefore not be fully used on the first day. In a practical example, the initial amount of capital is divided into, for example, 20 consecutive trading days. Subsequently, the closing of positions usually results in daily capital flows, which are used for further buy signals according to the described method.

The formulas with the aid of which buy and sell signals are generated in the securities analysis device according to the invention or the securities analysis method according to the invention are also called trading systems, which are created by adding logical conditions to indicators and chart histories.

All indicators have in common that they process at least one parameter from a series of five values each: - Starting course

- highest course

- lowest price

- closing price

- trading volume

This series of parameter values refers to an underlying cycle of, for example

- a trading day

- an hour - a quarter of an hour

- a minute or any other interval (up to a second), the length of which can be determined by the user, as far as his data provider this cycle can deliver. In the current version, the securities analysis device or the corresponding securities analysis method described in the present invention, inter alia, which is described as the World Market Screener, works with end-of-day values. Indicators and the trading systems based on them work with the stated values and link them by addition, subtraction, multiplication and division, whereby values are used over several cycles. In some cases, exponential functions and / or logarithmic functions are also used. In the specialist literature, indicators are divided into groups, each consisting of similarly constructed indicators. The division into groups is not uniform but depends on the respective author.

In the following, groups are described which are used by different authors in a relatively consistent manner and which are also used to categorize the indicators in the World Market Screener:

Moving average

The average is formed over several cycles, the cycle being advanced by one. Buy and sell signals arise when e.g. a longer and a shorter moving average intersect. Accumulation / Distribution

By evaluating the distance from the closing price to the high and low values in a cycle, it is determined whether there is a buying pressure or rather a selling pressure. The results are usually averaged over several cycles. The result of these indicators can e.g. used to support signal formation with other indicators.

oscillators

This is the largest group of indicators. Differences from a moving average create values that oscillate around zero or between a maximum and a minimum. Buy and sell signals arise in that the oscillator exceeds an upper threshold value or falls below a lower threshold value.

Trendermittlunq

By averaging over a longer period of time and then determining whether the resulting curve rises or falls, it is determined whether the market is in an upward or downward trend. The result is mainly used to verify the signals of other indicators, because some of them only deliver usable results in a certain type of market, e.g. can only be used with an upward trend, or with a downward trend or only in a sideways trend.

volatility

By evaluating the maximum and minimum values over several cycles, it is determined how large the fluctuation range of the market is. The resulting curve must be smoothed in order to obtain usable information. Declining volatility can then be used to increase the relevance of other indicators, because declining volatility indicates an impending market breakout in one direction or the other.

ratios

This type of indicator shows the relationship between two sizes, e.g. from performance to a market index or the average percentage change in performance when the market index changes by a certain amount. Zvklusindikatoren

A wave-like movement of the market is determined in the cycle indicators by forming a difference at a previous point in time and then averaging. Buy and sell signals then arise when an upper or lower limit is exceeded. market indicators

While all other indicators evaluate the values of only one security or the market index if necessary, the market indicators use a large number or all of the values traded on a stock exchange. The mood is often determined by counting securities with a stoned property, e.g. by summing all securities in the uptrend or in the downtrend and then forming the difference.

FIGS. 8 and 9 show the difference between previous security analysis methods and the security analysis method according to the invention. So far - as shown in FIG. 8 - it is customary to look at a chart history of an individual share by means of a trading system TS in the past tv and to draw conclusions from this in a linear view of the future t _{z of} the chart history. The triangle shown in the upper part in the middle of Fig. 9 illustrates that in the securities analysis method according to the invention, from a total of approximately 6000 selected shares, taking into account approximately 150 trading systems TS and taking into account the prevailing market conditions, a quantity of approx 550,000 trading variants TV are calculated per day. For this, a data processing device according to the invention with nine computer systems is required as the basis for the considerations made at the outset about the performance of today's computer systems. By filtering out the less successful trading variants in the test periods ti or t ₂ using a filter 520, this number is reduced to approximately 120,000 trading variants per day, which are added to a stored pool 530 of approximately 35 million trading variants. From this, the trading variants can be calculated within a computing time of less than 35 seconds, which corresponds relatively accurately to a chart's chronological history in the past. Experiments have shown that the chart movements of certain securities are highly likely to go through very similar cycles under the same or very similar market conditions. The match is graphically on the right lower part of Fig. 9 indicated by the dashed small triangle, which is placed over a chart of a particular security and finds a high degree of agreement there. This enables the trading variants to be determined that will provide roughly predictable results for certain securities under the given conditions.

The security analysis method shown in FIG. 9 or the security analysis device shown is not a rigid system or method, but can be adapted by changing parameters. For example, by actuating a first slider 500, an adjustment can be made to the behavior of the investor in such a way that the investor can switch between an investment form with a high risk based on "best profit" and an investment type oriented towards the "best ratio" with medium risk towards an investment form based on "best reliability" with the lowest risk. The terms "best ratio" and "best reliability" are sometimes used almost synonymously in the literature; in the present example, this is intended to mean an investment behavior with medium profit opportunities and medium risk.

Another preferably infinitely variable setting option is provided by a second slider 510 which reproduces the gradient angle of a vector which is the

Features market conditions. Depending on whether the market is rising

Has course that a long position L (corresponding to a slope angle of the

Vector of> 30 °) with a purchase of securities for a planned longer one

Period, or whether the market is declining, with a high trend to sell and a short position S (corresponding to a slope angle of the vector of <- 30 °), or whether the market between them is one

Sideways movement M with a slope angle of the vector between + 30 ° and

- Tends 30 °, the time parameters for the test times ti and t _{2 are} changed.

As a result, other trading variants that are advantageous than those that would be the case under different market conditions are determined in the calculation. The security analysis device according to the invention and the security analysis method according to the invention, in conjunction with the data processing device according to the invention and the data processing method according to the invention, revolutionize the possibilities of adapting both to investor behavior and to market conditions and based on a huge amount of data for the history of chart histories Selection of highly promising trading options.

Claims

claims

1. Data processing device for interactive multi-user access to large total amounts of data with a predetermined maximum response time, comprising a) a plurality of computer nodes (40) each equipped with its own working memory (44) and b) a load sharing and result synthesis device (20) for sharing at least one an interactive user request initiated data processing on the

A plurality of computer nodes (40) and for synthesizing an overall response to the interactive user request from individual partial responses supplied by the computer nodes (40); c) the total amount of data being broken down beforehand into disjoint subsets, each of which is completely stored in the working memory (44) of a computer node (40) uniquely assigned to it; d) the portion of the portion initiated by the interactive user request assigned to each computer node (40)

Data processing process is predetermined such that it is completed within a period of time that is less than or equal to a predetermined maximum duration; e) wherein each computer node (40) on the basis of the subset of the total amount of data assigned to it and stored in its working memory (44) independently of all other subsets of the working memory (44) of the other computer nodes (40) Total data volume generated its partial response.

2. Data processing device according to claim 1, characterized in that the following relation applies:

Z * D * B <F * P where: Z number of clock cycles each in the working memory

Computer node (40) stored subset of the total amount of data;

D number of subsets of the total amount of data to be processed for multi-user access; B Number of users who quasi-simultaneously

Retrieves multi-user access;

F number of processor cycles per second per computer node (40); and

P Number of computer nodes (40) used in parallel.

3. Data processing device according to claim 1 or 2, characterized in that the following relation applies:

S * D * B <V * P where: S memory requirement in the main memory (44) each

Computer node (40) stored subset of the total amount of data;

D Number of subsets of the total amount of data required for a

Multi-user access to be processed; B Number of users who quasi-simultaneously Retrieves multi-user access;

V speed of the main memory (44) when reading out data; such as

P Number of computer nodes (40) used in parallel.

4. Procedure for interactive multi-user access to large

Total amount of data with a given maximum response time

Execution on a plurality of each with its own

Computer memory (40) equipped with a working memory (44), comprising the following steps: a) breaking down the total amount of data into disjoint subsets, each of which is completely stored in the working memory (44) of a computer node (40) uniquely assigned to it; b) Splitting at least one by an interactive one

User request initiated data processing process on the plurality of computer nodes (40); c) Forming a partial response to the interactive user request on each computer node (40), ca), the one assigned to each computer node (40)

Portion of the data processing process initiated by the interactive user request is predetermined such that it is completely completed within a period of time that is less than or equal to a predetermined maximum duration; and cb) wherein each computer node (40) on the basis of the subset of the total amount of data assigned to it and stored in its working memory (44) generates its partial response independently of all other subsets of the total amount of data stored in the working memories (44) of the other computer nodes (40); and d) synthesis of an overall response to the interactive user request from individual partial answers provided by the computer nodes (40).

Securities analysis device for chart analysis of tradable securities, comprising a data processing device for interactive multi-user access to large amounts of total historical data

Security prices with a given maximum response time, the

Data processing device comprising a) a plurality of computer nodes (40) each equipped with its own working memory (44) and b) a load sharing and result synthesis device (20) for distributing at least one data processing process initiated by an interactive user request to the plurality of computer nodes (40) and for synthesizing an overall response to the interactive user request from individual ones supplied by the computer nodes (40)

Partial responses; c) wherein the total amount of historical stock price data is broken down beforehand into disjoint subsets, each of which is completely stored in the working memory (44) of a computer node (40) uniquely assigned to it; d) wherein the portion of the part initiated by the interactive user request that is assigned to each computer node (40) Data processing process is predetermined such that it is completed within a period of time that is less than or equal to a predetermined maximum duration; e) wherein each computer node (40) on the basis of the assigned to it and stored in its working memory (44) of the total amount of data of stock price data independent of all other subsets of the total amount of data stored in the working memories (44) of the other computing nodes (40) Partial response generated.

6. Security analysis device according to claim 5, characterized in that the security price data is divided such that the corresponding data of each security in each case

Main memory (44) of a single computer node (40) are stored.

7. The securities analysis device as claimed in claim 5 or 6, characterized in that price data is contained in the total amount of data for each security and trading day, each of which is shown as the difference between market values at different times on the same trading day.

8. The securities analysis device as claimed in claim 7, characterized in that the following price data are contained in the total amount of data for each security and trading day: a) difference value from the starting price and the lowest price, b) difference value from the closing price and the lowest price, and c) difference value from the maximum price and the closing price.

, Security analysis device according to claim 8, characterized in that the security data per security and trading day are each stored in a storage section with the following dimensions: a) 1 byte: exponent, namely

Bit 7-5: Common exponent to base 2 for display

aa) the difference between the starting price and the lowest price,

ab) the difference between the closing price and the lowest price, and

ac) the difference between the maximum price and the closing price;

Bit 4-2: exponent to base 10 for the closing price value,

Bit 1-0: exponent to base 10 for the value of the trading volume on the trading day;

b) 1 byte: mantissa of the difference value from the opening price and the lowest price, c) 1 byte: mantissa of the difference value from the closing price and the lowest price, d) 1 byte: mantissa of the difference value from the maximum price and

Closing price * e) 2 bytes: mantissa of the closing price, and f) 2 bytes: mantissa of the value of the trading volume on the trading day.

10. Security analysis method for interactive multi-user access Large total data volumes of historical securities prices with a predetermined maximum response time for execution on a plurality of computer nodes (40) each equipped with its own working memory (44), with the following steps: a) Decomposing the total data volume of historical ones

Securities prices into disjoint subsets, each of which is stored entirely in the working memory (44) of a computer node (20) uniquely assigned to it; b) splitting up at least one data processing process initiated by an interactive user request

Plurality of computer nodes (40); c) Forming a partial response to the interactive user request on each computer node (40), ca), the share of the interactive user request assigned to each computer node (40)

Data processing process initiated by the user request is predetermined such that it is completely completed within a period of time that is less than or equal to a predetermined maximum duration; and

cb) wherein each computer node (40) on the basis of the subset of the total amount of data assigned to it and stored in its working memory (44) is independent of all other subsets of the total amount of historical data stored in the working memories (44) of the other computing nodes (40)

Security prices generated its partial response; and

d) synthesis of an overall response to the interactive user request from individual partial responses supplied by the computer nodes (40).

11. The security analysis method according to claim 10, characterized in that the security price data is divided in such a way that the corresponding data of each security in each case

Main memory (44) of a single computer node (40) can be stored.

12. The security analysis method according to claim 10 or 11, characterized in that each security and trading day price data in the

Total amount of data are included, each as the difference of

Market values are shown at different times on the same trading day.

13. The securities analysis method according to claim 12, characterized in that the following price data in the total amount of data are processed per security and trading day:

a) difference between the starting price and the lowest price,

b) difference between the closing price and the lowest price, and

c) Difference between the maximum price and the closing price.

14. Security analysis device according to claim 13, characterized in that the security data per security and trading day are each stored in a storage section with the following dimensions: a) 1 byte: exponent, namely

Bit 7-5: Common exponent to base 2 for display

aa) the difference from the starting price and Lowest price,

ab) the difference between the closing price and the lowest price, and

ac) the difference between the maximum price and the closing price;

Bit 4-2: exponent to base 10 for the closing price value,

Bit 1, 0: base 10 exponent for the value of the trading volume on the trading day;

b) 1 byte: mantissa of the difference between the opening price and the lowest price,

c) 1 byte: mantissa of the difference between the closing price and the lowest price,

d) 1 byte: mantissa of the difference between the maximum price and the closing price,

e) 2 bytes: mantissa of the closing price, and

f) 2 bytes: mantissa of the value of the trading volume on the trading day.

15. Security analysis device according to claim 5, characterized in that by means of at least one mathematical formula (trading system) buy and sell signals of securities are generated on the basis of stored security prices, the formula processing at least one of the following input values: highest price, lowest

Price, closing price and trading volume, each related to one or more securities and to a defined time interval.

16. The securities analysis device as claimed in claim 15, characterized in that a multiplicity of different mathematical formulas are used for generating buy and sell signals, at least one of the formulas in each case belonging to one of the following eight groups: moving average, accumulation / distribution , Oscillators, trend determination, volatility, ratios, cycle indicators, market indicators.

17. Securities analysis device according to claim 16, characterized in that the buy and sell signals of mathematical formulas are evaluated with one another by means of a first slide controller (500) which, on the one hand, "Best ratio" between the positions "Best Profit" about in the middle and "Best reliability" on the other hand, is infinitely adjustable in order to optimize the yield and / or the risk of an investment in the security depending on investor behavior.

18. The securities analysis device according to claim 16, characterized in that the buy and sell signals of individual securities are linked to one another in such a way that a total amount of capital available to an investor is divided up for the purchase of several securities.

19. Security analysis method according to claim 10, characterized in that by means of at least one mathematical formula (trading system) buy and sell signals of securities on the basis of stored

Security prices are generated, the formula processing at least one of the following input values: starting price, highest price, lowest price, closing price and trading volume, each based on one or more securities and on a defined time interval.

20. The securities analysis method as claimed in claim 19, characterized in that a multiplicity of different mathematical formulas are used for generating buy and sell signals, at least one of the formulas in each case belonging to one of the following eight groups: moving average, accumulation / distribution , Oscillators, trend determination, volatility, ratio, cycle indicators, market indicators.

21. Securities analysis method according to claim 19 or 20, characterized in that the buy and sell signals of mathematical formulas are linked to one another by means of a slider which, between the positions "Best Profit" on the one hand, "Best ratio" approximately in the Middle and "Best reliability" on the other hand can be adjusted in stages to optimize the return and / or the risk of an investment in the security depending on investor behavior.

22. Securities analysis method according to one of claims 19 to 21, characterized in that the buy and sell signals of individual securities are linked to one another in such a way that a total amount of capital available to an investor is divided up for the purchase of several securities.

23. The securities analysis method as claimed in one of claims 19 to 22, characterized in that the mathematical formulas can be influenced by a position of a second slider (510) which on the one hand has the position "long", representing an increasing market, in has in the middle the position "Medium", representing a sideways market and on the other side the position "Short", representing a falling market, the position of the second slider (510) corresponding to the slope vector of the market behavior.

24. Securities analysis method according to one of claims 19 to 23, characterized in that one or more test periods (ti or t ₂ ) are used for the assessment and the final selection of trading systems.

25. The securities analysis method according to claim 24, characterized in that a first test period (ti) is used for establishing correlations between the behavior of trading systems and in that a second test period (t ₂ ) for determining currently successful trading Systems is used.

26. The securities analysis method according to claim 25, characterized in that correlations between the behavior of individual trading systems and an average trading system are calculated, which is formed from the currently successful trading systems.

27. Securities analysis method according to one of claims 24 to 26, characterized in that securities with a positive return or with a zero return in a shorter test period (t ₂ ) are selected for an application in custody account management in that their correlation with an average trading System of the currently successful trading systems is evaluated.