Patent Application
20080155015
The present invention relates to distribution of data in computer systems. In particular it relates to information dissemination in high performance trading systems.
Today it is very common that information is sent over computer networks. The amount of information being sent is rapidly increasing due to the advances in technology, making it possible to send and handle more information at higher speed compared to before. Furthermore, new applications demands a higher amount of information compared to before. Even further, the importance of information has open up a new field of business wherein information is sold.
An example of computer systems wherein information dissemination is one of the most essential parts are electronic trading systems.
Electronic trading of securities, derivatives, commodities and other financial instruments results in large amount of information which has to be distributed to users that needs the information for making trade decisions, statistical calculations and other assessments. Furthermore the users connected to such a centralised trading system want to have the information as soon as possible. In these cases it may not be enough to only boost the performance in the central system by for example updating the hardware, in order to get rid of a bottleneck or other latency problem in the system. Usually these bottlenecks end up at the user side anyway, since the users may have limited possibilities to update their connection to the central system.
Thus, this type of central system has to generate and distribute a lot of data to many different users not only on continous basis but also at specific times and occations during a trading day, and it has to be done in an efficient way.
An example of a system for providing electronic information is described in US 2005/0273421. This document describes a different system wherein the trading information and mulitiple types of electronic information is sent in the same data stream. Providing electronic trading information and electronic information on the same data stream significantly increases the computations required on a target device and by servers being used by electronic trading providers to separate the information. The system disclosed in us 2005/0273421 solves this problem by splitting the first data stream into plural second data streams that can be selectively requested, displayed and used by a user. Each of the plural second data streams includes one or more of the plural different types of electronic trading information from the first data stream. Thereby allowing an individual target device to selectively request, receive and use the one or more of the plural types of electronic trading information in the second data stream faster than using the same electronic trading information from the first data stream.
However, this type of filtering is often much to rough since a user/client is most likely interested in other data streams too. Thus the above described system has drawbacks, especially in environments having limited bandwidth connection, since it splitts the information based on the type of electronic information.
Hence the above system is useless in networks where the data rates of the connections for users is limited and varies between different users.
Some trading system has other functionalities for distributing information. However, the problem for these solutions is that they require the system to generate and send duplicated information. These solutions may for example generate one message flow for a user A (with a great connection) that contains the entire depth and one message flow for a user B (with a poor connection) that contains a limited picture of the market (top X levels). The two messages will therefore contain duplicated information for the top X levels. This will result in extra load on both the central systems performance and the traffic on the central and peripheral network.
Another problem with electronic information systems is that usually there is a central processor/s that aggregates the information and distributes it. Since the processor has to use time on aggregating and spreading duplicated information there will be less time for other tasks such as receiving information and thus bottlenecks may occur in other parts of the system.
Another problem with the prior art systems is that they do not consider counter performances provided by customers when distributing data. Thus some customers may expericence an unfair treatment.
Thus it is an object of the present invention to provide a solution for disseminating data in a more efficient way.
It is another object of the present invention to provide a solution for sending out data to users having different data rate of their data connections.
It is another object of the present invention to provide a solution for sending out data in a bandwidth efficient way.
It is another object of the present invention to provide a solution for sending out data without the need of duplicating the data.
It is another object of the present invention to consider customer counter performances when sending out data.
According to a first aspect of the invention the above and other objects are achieved by a computer system for sending data sets to a plurality of clients via a network, the computer system comprising:
The above computer system has the advantage that it provides a solution that makes it possible to disseminate data in a more efficient way. The system makes it possible to send data using less bandwidth (data rate) and performance in the computer system since duplications of messages are preferably not extracted or sent.
The devices in the system are connected to each other by interfaces. There are two types of interfaces; hardware interfaces and software interfaces.
Hardware interfaces are plugs, sockets, cables and electrical signals traveling through them. Examples are USB, FireWire, Ethernet, ATA/IDE, SCSI, PCI and so forth.
Software interfaces (programming interfaces) are the languages, codes and messages that programs use to communicate with each other and to the hardware. Examples are the Windows, Mac and Linux operating systems, SMTP e-mail, IP network protocols and the software drivers that activate the peripheral devices.
Both types of interfaces are needed in order to get the internal and external devices in a computer system/network to communicate with each other.
The term data element may be referred to as a named unit of data that, in some contexts, is considered indivisible and in other contexts may consist of data items. In this document the data element can either be considered indivisible or it may consists of data items. Preferably the data elements in this document are sorted according to ranking rules. For example the data elements could be sorted by ranking rules such as: price, time, price and time, traders before market makers, inverted price or sorted by volume and so forth, or any combination of the mentioned ranking rules.
The superset may comprise a number of data elements that are to be distributed to a plurality of clients. Preferably the superset of data elements is stored in a memory such as in the primary memory e.g. the cash memory, RAM, DRAM and so forth, or in a secondary memory such as hard disk, USB memory or the alike. However since the primary memories are much faster compared to the secondary memories the primary memories are better suited for the task.
The data elements may comprise data relating to orders in an order book. An order is typically an instruction issued by a client. Usually the order is a buy or a sell order for a certain or a plurality of financial instruments. However there exist many different types of orders, therefore in this document orders are not only limited to buy or sell orders.
Example of preferred content of a data element is at least one of the following data: trader identity, series, bid or ask, price, quantity, account or client. The data element may also include other fields such as free text field and so forth.
The subsets may comprise extracted parts of the superset of data. Preferably the subsets do not overlap each other with regard to data elements. The superset can therefore be represented by a number of subsets.
How many data elements each subset should comprise may be chosen by a system administrator at the central computer system. For example the subset representing the first part of the superset may comprise five levels of data elements, the subset representing the second part of the superset may comprise the following ten levels of data elements and the subset representing the third part of the superset may comprise the remaining data elements in the superset. This is just an example of one embodiment, in other embodiments more subsets comprising less data elements may be used in order to represent the superset, or less subsets comprising more data elements.
In a further embodiment of the invention it is possible for a client to subscribe to subsets comprising a number of data elements chosen by the client. Hence it would be possible to tailor the size of the subsets to individual customer needs. The size and number of subsets could also be based on a counter performance of a customer, such as payment, number of trades and so forth. More examples are mentioned below in the text.
The selector in the computer system preferably selects a group of clients based on at least one of the following:
In this way the dissemination of data can be done in an intelligent/selective way since the computer system is able to detect a data rate on a connection to a client. This may be done by measuring the data rate on the data connection or by information received from the client, or by information from a message bus and so forth. Thus the selector may select a group of clients based on a combination of data rate and subscription or based on one of them alone. In another situation the selector may select a group of clients based on an infrastructure load.
The input regarding data rate or “infrastructure load” may be obtained by measuring the data rate on the data connection, processor load at the central system, or by information related to an application program received from a client computer, or by information from a bus such as an intelligent message bus with ability to measure current bandwidth situation at different locations by use of network surveillance tools. Hence the infrastructure load may be related to any activity in the market on an electronic exchange.
Furthermore, it may also be possible for a client to provide change-levels to the central system regarding the data rate on the connection. Thus if the measured data rate on a connection exceeds an upper change-level the connection is busy and less information can be sent via the connection. If the measured data rate on the connection falls below a lower change-level the connection is not busy and more information can be sent. By providing this functionality the system can always make sure that a certain client receives the best service of information based on the data rate.
It is also possible for the central system to control its information dissemination based on the above so that orders sent from the client is not prevented from entering the central system. Thus if the measured data rate on a connection to a client is high the central system may decide to wait or cancel some of the information that will be sent to the client. By providing this functionality the system can always make sure that a certain client's orders/request into the central system will arrive faster since the connection is not busy with other information sent from the exchange.
Preferably the superset of sorted data elements comprises data relating to at least one of the following data:
However, data in any sorted list may be used in relation to the present invention and not only the data mentioned above. The above may be used where it is hard for an external application to sort the incoming data. By sending the data already sorted it speeds up the processing of the received data for a client computer.
Furthermore the data elements in the subsets of data comprises at least one of the following:
By sending the delta-changes valuable bandwidth can be saved. If only one part of a data element in the superset has changed it may be better to only send the delta change of that part in order to update the data set at the client computer. However in other situations it may be more efficient to send one or more data elements to replace obsolete data elements at the client. In these cases a snapshot of the superset of data can be sent and the amount of bandwidth to be used becomes easier to calculate and predict.
By sending the changes using delta changes it is possible to keep a list at a client computer updated in real time without the need of providing the whole list each time. Furthermore little bandwidth may be used each time for each message. However logic is necessary at the client computer since the client has to be able to perform implicit actions/calculations on the data received. Preferably used for long lists with few changes/updates.
Sending the changes by snapshots makes it is easier for a client computer since it is not necessary for the client computer to apply any logic on the data received. The client computer only replaces the old message with the new message and no further implicit calculation/action is usually necessary. Furthermore it is possible to limit the maximum bandwidth by hold-back timers, however more bandwidth is necessary for each message. Preferably used for short lists with many updates/changes.
The memory mentioned above may further comprise an area, accessible by the selector for storing subscriptions associated with clients. By having this feature it becomes easier for the selector to select groups of clients since the subscription may comprise information such as: customer needs, the connections data rate, number of trades, the client's turnover, client status such as market maker and so forth.
The subscription associated with a client is preferably based on at least one of the following:
Since clients usually have to pay for the subscriptions of information, the service or quality level of the subscriptions varies. For example the more a client pays the more information the client can obtain. Thus as mentioned above the subscription may be associated with a client based on the connection the client has. The client will get a subscription suitable for the data rate on the connection. The reason for this is that the client may be restricted by the connection so that the client can only receive a certain amount of data. However the subscription may also be based on a counter performance from the client. In this case the connection to the client does normally not constitute a bottleneck with regards to data-rate. Thus depending on the counter performance the subscription associated to the client may change. Example of counter performances can be: payments, or turnover at the clients account, liquidity provided by the client, number of trades the client has done, data received from the client and so forth. Hence, the more the client trades he will be upgraded to a subscription with a better service or quality.
The clients may be associated with specific rights, “client rights”. For example a market maker may have some privileges/rights compared to other clients. Thus some information could be associated with specific rights and thereby only sent to the clients having that right/privilege.
The data rate of the connection to the client can be measured in many different ways. A few examples (as mentioned earlier) are; by measuring the data rate on the connection and/or by information provided by the client and/or information provided by a data message bus using surveillance tools known to the person skilled in the art.
The extractor preferably extracts a subset of data elements from the superset of sorted data elements upon at least one of the following events:
When to extract and/or send a subset may be chosen depending on the situation. If changes in the superset do not occur so often it may be more efficient to extract data elements from the superset upon a change in the superset. Another solution is to extract data elements at certain time intervals decided by a holdback timer. In these cases it is easier to predict bandwidth utilization in the system.
Furthermore data elements may be extracted by a combination of the two above. For example, as a default, data elements could be extracted at certain time intervals but if no change in the superset occurs there is no need to send any information. Thus the systems checks if any change has occurred and if no change has occurred in the superset no data elements are extracted or sent not even at certain time intervals. Not until a change is detected by the system, then the system starts sending data elements at certain time intervals until no changes in the superset is detected again. The system waits until a change occurs and then starts sending data elements again.
Therefore, the computer system may furthermore comprise a comparator associated with the memory, for detecting at least one location of a change in the superset of sorted data elements, and based on the location sending a subset associated with the location to clients subscribing for the subset. Furthermore changes in adjacent subsets which occur as a consequence of the change in the superset are preferably added to the subset being sent. Thus the subset being sent preferably comprises an additional data element, originally associated with another subset, if the change was e.g. a deletion of a data element in the superset.
The memory in the computer system may further comprise a second superset of sorted data elements; the data elements in the second superset preferably comprise one or more of the data elements from the first superset.
Thus two supersets may be used for extracting and sending data to the clients. The second superset is preferably treated the same way as the first superset. However since the second superset may have data elements sorted in a different way changes may occur at other positions in the superset.
In a second aspect of the invention, the above and other objects are fulfilled by an electronic exchange comprising the computer system as mentioned above.
The computer system may thus be an integrated module in an electronic exchange. It can also be a stand alone module that can be sold separately as an information dissemination system.
In a third aspect of the invention, the above and other objects are fulfilled by a method implemented on a computer system for sending data sets to a plurality of clients via a network, the method comprising the steps of:
The above method has the advantage that it provides a solution that makes it possible to disseminate data in a more efficient way. The method makes it possible to send data without affecting bandwidth and performance in the computer system since duplications of messages are preferably not extracted or sent.
The method may further comprise the step of selecting groups of clients, based on at least one of the following input:
In this way the dissemination of data can be done in a smart way since the method makes it possible to sense a data rate on a connection to a client. For example this may be done by measuring the data rate directly on the connection, or based on information from the client, another solution could be to receive information from a message bus and so forth. Thus the selector may select a group based on a combination of data rate and subscription or based on one of them alone.
The method may further comprise the step of arranging the data in the subsets according to at least one of the following arrangements:
By arranging the data in the subsets valuable bandwidth can be saved. If only one part of a data element in the superset has changed it may be better to only send the delta change of that part in order to update the data set at the client computer. However in other situations it may be more efficient to send one or more data elements to replace obsolete data elements at the client. In these cases a snapshot of the superset of data can be sent and the amount of bandwidth to be used becomes easier to calculate and predict.
In another embodiment the method may comprise the step of associating the subscription with a selected client based on at least one of the following input:
The extracting step, preferably extracts the subsets from the superset of data elements upon at least one of the following events:
When and how the subset is extracted may be chosen depending on the specific situation. For example if changes in the superset do not occur so often, it may be more efficient to extract data elements from the superset upon a change in the superset. However if changes continuously occurs at relatively high speed the solution may be to extract data elements (snapshot) at certain time intervals decided by a holdback timer. In these cases it is easier to predict bandwidth utilization in the system and it may be more bandwidth efficient to do it this way.
Thus the system may chose to send a snapshot of a part of the dataset. When using snapshots a holdback timer may be used in order to decide when a snapshot should be sent. For example a snapshot of the four uppermost data element is sent every second. The size of this message will always be the same (e.g. 4 bytes, 1 bytes per element). Another solution would be to send a data element for each update when it occurs. If during this second 100 updates occur this would result in 100 data elements a 1 byte, being sent. Thus 100 bytes is sent during the interval of one second. From a bandwidth perspective it may thus be better to use the snapshot and timer technique in this situation.
As described earlier in this document, data elements may be extracted by a combination of the two techniques, upon a change and upon time intervals.
The method may further comprise the following step:
This has the advantage that only the subset comprising the change in the superset will be extracted and sent to a client. In this way bandwidth and other system resources such as processor time will be saved.
The method may also comprise the following step:
This has the advantage that processing of the data elements can be done at the client computer since a deletion/removal of a data element in a first subset means that the first data element in the subset following the first subset has to be moved to the first subset.
Preferably the subsets of data comprise data relating to at least one of the following data:
As mentioned earlier in this document the above data may be used where it is hard for an external application to sort the incoming data. By sending the data already sorted it speeds up the processing of the received data for a client computer.
In a fourth aspect of the invention, the above and other objects are fulfilled by a client computer for management of sorted data elements, the client computer comprising:
The above client computer has the advantage that it provides a solution that makes it possible to process data received from a computer system in a more efficient way. Since the sorted data elements in the memory needs to be updated at certain time intervals so that a user of the computer can see what is happening at a central computer system such as in an order book, the client computer comprises a message handler that treats the data elements based on the information received in the update message. Thus if the client computer receives an update message comprising delta changes it adds the delta changes to the data elements stored in the memory. If the client computer receives an update message comprising data elements it may replace the obsolete data elements in the memory in order to provide an accurate and correct data to a user of the client computer.
The client computer may further comprise a management module for executing implicit operations on the stored set of sorted data elements based on the data in the update message. The implicit operations preferably relates to at least one of the following operations: push operation, pop operation, delete operation and so forth.
The implicit operations are executed by the client computer on the local dataset stored in the memory of the client computer. The implicit operation is based on the content of the update message in order to create a sorted dataset at the client computer.
The data in the update message may be chosen from the group of data comprising:
Depending on the data the client computer receives in the update message the client computer performs certain operations as described above.
In a fifth embodiment of the invention, the above and other objects are fulfilled by a computer program product according to any of the previous described embodiment, the computer program product being stored on a data carrier.
These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.
In this application some specific terms are used, below follows a description of these.
MBO—Market By Order—In
MBP—Market By Price—This type of message 11 contains information on the price levels that exists in the market with a summary of the total volume on each level. It is possible to say that only the top Y price levels will be distributed.
A MBO example will be used as support for the description, in order to simplify the understanding of the invention. In this example the system is configured to distribute three different levels of MBO, in order to support different type of bandwidth users. However the invention is useful with any type of sorted lists and not only with MBO or MBP message information.
The invention will now be explained in detail with reference to the figures.
In the following the invention will be described in detail with reference to the figures. The details of how the subsets may be sent to the client computers and how a client computer preferably manages the data received in the update messages sent from the central system, will be explained.
In the example provided in this document there exist three clients; A, B and C having different data rates on the connections as explained below.
Client A has a limited connection against the central system with poor bandwidth. He can therefore not subscribe for the entire order depth. Therefore he only wants to subscribe for the top 5 orders in the depth.
Client B is a standard member client who is trading on one of the big firms 5 and has normal connection to the central system 2. He wants to subscribe for the top 20 orders.
Client C is a high standard client that is connected to a central gateway 3 and has a high performance connection. He wants to subscribe for the entire order depth.
Below follows an example of how the invention preferably works when disseminating information from a superset 7 of data.
Either the clients themselves sets up a subscription. However this may also be done centrally at the central computer system 2, by a system administrator or automatically by the system itself based on a number of predetermined parameters such as data rate and so forth.
The central system 2 will now have a method of distributing order book changes so that it supports different type of bandwidth clients. The trick of this is to distribute this information without the need of duplicating messages for different type of clients.
The orderbook 7 contains 30 orders on the offer side. The first subset 8 has been sent out for the top 5 orders. The second subset 9 has been sent out for the orders on position 6 to 20 and the third subset 10 has been sent out for the remaining levels.
Client A will therefore have a correct picture of the top 5 orders. Client B will have correct picture of the top 20 orders and finally client C will have a full order depth picture. See
The following section will explain how an add order shall be treated within the system to support this type of solution. It is based on a solution where order book changes are distributed as delta changes. The example is illustrated in
Add
Level 1-5—A new order with a ranking position between 1 and 5.
The system according to the invention will generate the first subset Add order 26 broadcast. All clients will receive this broadcast 26.
In
In
In
The following examples of sending a subset add order message for the second and third subset is not picturized in the figures. However the same method applies as described in the example above.
Level 6-20—A new order with a ranking position between 6 and 20.
The system according to the invention will generate the second subset Add order broadcast. Only Client B and Client C will receive this broadcast. Client A will not receive this message and his network connection against the central system will therefore not be affected by this change.
Client B will insert the order and update the ranking position for all other orders that are affected by this change. The order that had position 20 and now has position 21 will be removed by the front-end application since the client no longer will receive any order changes for this order since he doesn't subscribe for the third subset 10 message.
Client C will insert the order and update the ranking position for all other orders that are affected by this change. The front-end application does not need to delete any order since he has a full order depth view by subscribing for all three broadcasts.
Level 21->—A new order with a ranking position >21.
The system according to the invention will generate the third subset Add order broadcast. Only Client C will receive this broadcast. Client C will insert the order and update the ranking position for all other orders that are affected by this change. The front-end application does not need to delete any order since he has a full order-depth view by subscribing for all three broadcasts.
Client A and Client B will not receive this message and their network connection against the central system will therefore not be affected by this change.
Remove:
The following example is picturized in the
Level 1-5—An order with a ranking position X between 1 and 5 is removed.
The system according to the invention will generate the first subset remove order broadcast 27. All clients will receive this broadcast 27. The broadcast 27 preferably contains 2 items. One remove on position X and one add on position 5. The Add on position 5 will contain the order that moved from position 6 to 5 as a result of the deleted order. The system will also generate a second subset add order 28 on position 20 for the order that moved from position 21 to 20. Another way of handling this would be to have an extra add record in the first subset message 27 for position 20.
Client A will remove the order on position X and improve the ranking position for all other orders that are affected by this change. The order that had position 4 will get position 3 and so on. Client A will also use the Add order in the message and insert it to the 5th position. Client A will not receive the second subset order broadcast.
Client B will remove the order on position X and improve the ranking position for all other orders that are affected by this change. Client B's application will ignore the order Add on position 5 since he already has this order in his order depth (via the subscription on the second subset 9). However, he may also receive the second subset order 28. This broadcast contains an Add order record on position 20. Client B's application will Add this order to the 20th position.
Client C will remove the order on position X and improve the ranking position for all other orders that are affected by this change. Client C's application will ignore the order Add on position 5 since he already has this order in his order depth (via the subscription on the second subset 9). He will also ignore the second subset order broadcast since it contains an order that he already has within his depth via his subscription on third subset 10.
The following examples of sending a subset remove order message for the second and third subset is not picturized in the figures. However the same method applies as described in the example above.
Level 6-20—An order with a ranking position X between 6 and 20 is removed. The system according to the invention will generate the second subset remove order broadcast. Only Client B and Client C will receive this broadcast. The broadcast will contain 2 items. One remove on position X and one add on position 20. The Add on position 20 will contain the order that moved from position 21 to 20 as a result of the deleted order.
Client A will not receive this message and his network connection against the central system will therefore not be affected by this change.
Client B will remove the order on position X and improve the ranking position for all other orders that are affected by this change. The order that had position 15 will get position 14 and so on. Client B will also use the Add order in the message and insert it to the 20th position.
Client C will remove the order on position X and improve the ranking position for all other orders that are affected by this change. Client C's application will ignore the order Add on position 20 since he already has this order in his order depth via the subscription on the third subset.
Level 20—An order with a ranking position X>20 is removed.
The system according to the invention will generate the third subset remove order broadcast. Client C will receive this broadcast. The broadcast will contain one remove item on position X.
Neither Client A nor Client B will receive this message and their network connections against the central system will therefore not be affected by this change.
Client C will remove the order on position X and improve the ranking position for all other orders that are affected by this change.
Update:
Below follows the explanation on how an order update shall be treated within the system to support this type of solution. It is based on a solution where order book changes are distributed as delta changes.
An update occurs on an existing order in the order book. This can happen when an order is partly traded or modified without any order book priority change. The solution is picturized in the
Level 1-5—Update that affects an existing order within the top 5 levels. The system according to the invention will generate the first subset order change 30 broadcast. All clients will receive this broadcast and update their order depth accordingly.
Level 6-20—Update that affects an existing order within the levels 6-20.
The system according to the invention generates the second subset change order 30 broadcast. Only Client B and Client C will receive this broadcast and update their order depth accordingly, see position 10 in the figures. Client A will not receive this message and his network connection against the central system will therefore not be affected by this change.
Level 21->—Update that affects an existing order the levels 21 and above.
The system according to the invention generates the third subset change order 30 broadcast. Only Client C will receive this broadcast and update his order depth accordingly, see position >21 in
The information dissemination method that is described in this document can be used for many different types of data. The solution can be used in all cases where you have a list of items and where there are clients that are interested in different parts of the list and you would like to distribute the list in real time in an efficient way without duplicating any messages.
In the above description the term “comprising” does not exclude other elements or steps and “a” or “an” does not exclude a plurality.
Furthermore the terms “include” and “contain” does not exclude other elements or steps.
