System and method for low latency market data

• US 8,130,758 B2
• Filed: 04/24/2006
• Issued: 03/06/2012
• Est. Priority Date: 06/27/2005
• Status: Active Grant

• Current Assignee: Bank of America Corp.
• Original Assignee: Bank of America Corp.
• Inventors: Cohen, David
• Primary Examiner(s): Wilson, Robert
• Assistant Examiner(s): Zhao, Wei
• Application Number:
  US11/410,536
• Publication Number:
  US 20070025351A1
• Time in Patent Office: 2,143 Days
• Field of Search: 370/403, 370/442, 725/110, 725/114, 709/246
• US Class Current: 370/392
• CPC Class Codes: G06F 2209/544 Remote
• CPC Class Codes: G06F 9/542 Event management; Broadcasting; Multicasting; Notifications
• CPC Class Codes: G06Q 40/04 Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
• CPC Class Codes: H04L 67/55 Push-based network services

Abstract

A method for low latency data comprises receiving a data stream comprising data from a data source, receiving the data via one or more queues, where each of the one or more queues is associated with one or more filters adapted to filter the data based at least upon a time predicate to create filtered data. Each of the one or more filters is established by one or more subscribers. The method also comprises receiving a portion of the filtered data that was filtered by a filter established by a particular one or more subscribers, and publishing the portion of the filtered data for the particular one or more subscribers.

Claims

1. An apparatus for low latency data, the apparatus comprising:

one or more feed handlers, each of the one or more feed handlers configured to;

receive a stream of market data, the market data comprising a plurality of market data segments; and

extract the plurality Of market data segments from the received stream of market data;

one or more queues comprising one or more data partitions, the one or more queues coupled to the one or more feed handlers via a direct connection, the direct connection comprising a cluster interconnect of Beowulf style, the one or more queues configured to;

receive the plurality of extracted market data segments from the one or more feed handlers;

store each of the extracted market data segments in a particular one of the one or more data partitions based on a characteristic of each of the extracted market data segments; and

filter the plurality of extracted market data segments stored in the one or more data partitions using one or more filters established by a particular one or more subscribers to create filtered data; and

one or more publishers, each of the one or more publishers configured to;

receive a portion of the filtered data;

publish the portion of the filtered data for the particular one or more subscribers; and

send, via a callback mechanism, a notification to the particular one or more subscribers when the one or more publishers receives the portion of the filtered data that was filtered by the filter established by the particular one or more subscribers;

the one or more queues operable to filter the plurality of extracted market data segments stored in the one or more data partitions by creating a plurality of jobs and scheduling the plurality of jobs to be executed in parallel on one or more computer systems in a computational cluster; and

wherein publishing the portion of the filtered data for the particular one or more subscribers comprises;

determining whether the filtered data is associated with a shared subscription;

publishing the filtered data to the particular one or more subscribers using an IP multicast channel if it is determined that the filtered data is associated with a shared subscription; and

publishing the filtered data to the particular one or more subscribers using a point-to-point messaging transport if it is determined that the filtered data is not associated with a shared subscription.

2. The apparatus of claim 1 wherein the particular one or more subscribers comprises an IP multicast group.

3. The apparatus of claim 1 wherein the particular one or more subscribers comprises a plurality of subscribers.

4. The apparatus of claim 1 wherein the one or more queues and the one or more publishers comprise a plurality of independent processes coupled by an asynchronous communication mechanism.

5. The apparatus of claim 4 wherein each of the plurality of independent processes are assigned to a cluster partition.

6. The apparatus of claim 5 wherein the cluster partition is configured with cluster resources based at least in part on an advance reservation facility.

7. The apparatus of claim 1 wherein the feed handler is configured to receive one or more direct feeds from one or more markets.

8. The apparatus of claim 1 wherein the apparatus is further configured to maximize data parallelism using a farmer/worker programming pattern.

9. The apparatus of claim 1 wherein the apparatus is further configured to maximize data parallelism using a data parallelism programming pattern.

10. The apparatus of claim 1 wherein the apparatus is further configured to maximize data parallelism using a functional/task programming pattern.

11. The apparatus of claim 1 wherein the apparatus is further configured to provide value-added analytics.

12. The apparatus of claim 11 wherein the apparatus is further configured to provide the value-added analytics by:

applying a new algorithm to a metric calculation; and

making a resulting stream available to one or more subscribers.

13. The apparatus of claim 12 wherein the metric comprises VWAP (Volume Weighted Average Price).

14. The apparatus of claim 12 wherein the apparatus is further configured to receive one or more subscriptions from the one or more subscribers, the one or more subscriptions expressed as an SQL-like query.

15. The apparatus of claim 14 wherein the one or more subscriptions are issued as continuous data stream queries which provide an asynchronous callback method.

16. The apparatus of claim 1 wherein the one or more publishers are further configured to invoke a callback method provided by the particular one or more subscribers when the one or more publishers receives the portion of the filtered market that was filtered by the filter established by the particular one or more subscribers.

17. The apparatus of claim 15 wherein the asynchronous callback method is implemented using one or more of Linux Asynchronous 110 (AI0), .NET Delegate and J2EE Message-Driven Beans (MDB).

18. The apparatus of claim 15 wherein the apparatus is further configured to maximize data parallelism, the task parallelism implemented using one or more of C/C++ using Asynchronous I/O (AI0), .NET Delegate, and J2EE.

19. The apparatus of claim 1 wherein a StreamBase™

stream-processing engine comprises the one or more queues and the one or more publishers.

20. The apparatus of claim 14 wherein the apparatus is further configured to register the one or more subscriptions using a remote procedure call (RPC) mechanism that includes a partition identifier, the apparatus further configured to return a subscription ID for subsequent use in subscription management.

21. The apparatus of claim 1 wherein the apparatus resides at a root of a sensor network.

22. The apparatus of claim 1 wherein the apparatus is further configured to allocate costs to one or more consuming applications associated with the one or more subscribers, the allocation of costs being based at least in part on queries submitted by the one or more subscribers, and on the underlying resources consumed by the queries.

23. The apparatus of claim 1 wherein the one or more feed handlers, the one or more queues, and the one or more publishers share a single monotonic clock, the clock for use in synchronizing a sampling interval across the one or more queues.

68. The apparatus of claim 1, wherein storing each of the extracted market data segments in a particular one of the one or more data partitions based on a characteristic of each of the extracted market data segments comprises storing each of the extracted market data segments in a particular one of the one or more data partitions based on an exchange symbol in the market data segment.

69. The apparatus of claim 1, wherein the one or more filters comprise at least a time predicate.

24. A method for low latency data, the method comprising:

receiving a stream of market data, the market data comprising a plurality of market data segments;

extracting the plurality of market data segments from the received stream of market data;

receiving the plurality of extracted market data segments via one or more queues, the one or more queues comprising one or more data partitions, the plurality of extracted market data segments received via a direct connection, the direct connection comprising a cluster interconnect of Beowulf style;

storing each of the extracted market data segments in a particular one of the one or more data partitions based on a characteristic of each of the extracted market data segments;

filtering the plurality of extracted market data segments stored in the one or more data partitions using one or more filters established by a particular one or more subscribers to create filtered data, the filtering performed by creating a plurality of jobs and scheduling the plurality of jobs to be executed in parallel on one or more computer systems in a computational cluster;

receiving a portion of the filtered data; and

publishing the portion of the filtered data for the particular one or more subscribers, wherein publishing the portion of the filtered data for the particular one or more subscribers comprises;

determining whether the filtered data is associated with a shared subscription;

publishing the filtered data to the particular one or more subscribers using an IP multicast channel if it is determined that the filtered data is associated with a shared subscription; and

publishing the filtered data to the particular one or more subscribers using a point-to-point messaging transport if it is determined that the filtered data is not associated with a shared subscription; and

sending, via a callback mechanism, a notification to the particular one or more subscribers when the portion of the filtered data that was filtered by the filter established by the particular one or more subscribers is received.

25. The method of claim 24 wherein the particular one or more subscribers comprises an IP multicast group.

26. The method of claim 24 wherein the particular one or more subscribers comprises a plurality of subscribers.

27. The method of claim 24 wherein the one or more queues and one or more publishers comprise a plurality of independent processes coupled by an asynchronous communication mechanism.

28. The method of claim 27 wherein each of the plurality of independent processes are assigned to a cluster partition.

29. The method of claim 28 wherein the cluster partition is configured with cluster resources based at least in part on an advance reservation facility.

30. The method of claim 24 wherein the feed handler is configured to receive one or more direct feeds from one or more markets.

31. The method of claim 24, further comprising maximizing data parallelism using a fanner/worker programming pattern.

32. The method of claim 24, further comprising maximizing data parallelism using a data parallelism programming pattern.

33. The method of claim 24, further comprising maximizing data parallelism using a functional/task programming pattern.

34. The method of claim 24, further comprising providing value-added analytics.

35. The method of claim 34, further comprising providing the value-added analytics by:

applying a new algorithm to a metric calculation; and

making the resulting stream available to one or more subscribers.

36. The method of claim 35 wherein the metric comprises VWAP (Volume Weighted Average Price).

37. The method of claim 35, further comprising receiving one or more subscriptions from the one or more subscribers, the one or more subscriptions expressed as an SQL-like query.

38. The method of claim 37, further comprising issuing the one or more subscriptions as continuous data stream queries which provide an asynchronous callback method.

39. The method of claim 24, further comprising invoking a callback method provided by the particular one or more subscribers when the one or more publishers receives the portion of the filtered market that was filtered by the filter established by the particular one or more subscribers.

40. The method of claim 38 wherein the asynchronous callback method is implemented using one or more of Linux Asynchronous 110 (A10), .NET Delegate and J2EE Message-Driven Beans (MDB).

41. The method of claim 38, further comprising maximizing data parallelism using one or more of C/C++ using Asynchronous I/O (A10), .NET Delegate, and J2EE.

42. The method of claim 24 wherein a StreamBase™

stream-processing engine comprises the one or more queues and the one or more publishers.

43. The method of claim 37, further comprising:

registering the one or more subscriptions using a remote procedure call (RPC) mechanism that includes a partition identifier; and

returning a subscription ID for subsequent use in subscription management.

44. The method of claim 24, further comprising allocating costs to one or more consuming applications associated with the one or more subscribers, the allocation of costs being based at least in part on queries submitted by the one or more subscribers, and on the underlying resources consumed by the queries.

70. The method of claim 24, wherein storing each of the extracted market data segments in a particular one of the one or more data partitions based on a characteristic of each of the extracted market data segments comprises storing each of the extracted market data segments in a particular one of the one or more data partitions based on an exchange symbol in the market data segment.

71. The method of claim 24, wherein the one or more filters comprise at least a time predicate.

45. A program storage device readable by a machine, embodying a program of instructions executable by the machine to perform a method for low latency data, the method comprising:

receiving a stream of market data, the market data comprising a plurality of market data segments;

extracting the plurality of market data segments from the received stream of market data;

receiving the plurality of extracted market data segments via one or more queues, the one or more queues comprising one or more data partitions, the plurality of extracted market data segments received via a direct connection, the direct connection comprising a cluster interconnect of Beowulf style;

storing each of the extracted market data segments in a particular one of the one or more data partitions based on a characteristic of each of the extracted market data segments;

filtering the plurality of extracted market data segments stored in the one or more data partitions using one or more filters established by a particular one or more subscribers to create filtered data, the filtering performed by creating a plurality of jobs and scheduling the plurality of jobs to be executed in parallel on one or more computer systems in a computational cluster;

receiving a portion of the filtered data; and

publishing the portion of the filtered data for the particular one or more subscribers, wherein publishing the portion of the filtered data for the particular one or more subscribers comprises;

determining whether the filtered data is associated with a shared subscription;

publishing the filtered data to the particular one or more subscribers using an IP multicast channel if it is determined that the filtered data is associated with a shared subscription; and

publishing the filtered data to the particular one or more subscribers using a point-to-point messaging transport if it is determined that the filtered data is not associated with a shared subscription; and

sending, via a callback mechanism, a notification to the particular one or more subscribers when the portion of the filtered data that was filtered by the filter established by the particular one or more subscribers is received.

46. The program storage device of claim 45 wherein the particular one or more subscribers comprises an IP multicast group.

47. The program storage device of claim 45 wherein the particular one or more subscribers comprises a plurality of subscribers.

48. The program storage device of claim 45 wherein the one or more queues and the one or more publishers comprise a plurality of independent processes coupled by an asynchronous communication mechanism.

49. The program storage device of claim 48 wherein each of the plurality of independent processes are assigned to a cluster partition.

50. The program storage device of claim 49 wherein the cluster partition is configured with cluster resources based at least in part on an advance reservation facility.

51. The program storage device of claim 45 wherein the feed handler is configured to receive one or more direct feeds from one or more markets.

52. The program storage device of claim 45, the method further comprising maximizing data parallelism using a farmer/worker programming pattern.

53. The program storage device of claim 45, the method further comprising maximizing data parallelism using a data parallelism programming pattern.

54. The program storage device of claim 45, the method further comprising maximizing data parallelism using a functional/task programming pattern.

55. The program storage device of claim 45, the method further comprising providing value-added analytics.

56. The program storage device of claim 55, the method further comprising providing the value-added analytics by:

applying a new algorithm to a metric calculation; and

making the resulting stream available to one or more subscribers.

57. The program storage device of claim 55 wherein the metric comprises VWAP (Volume Weighted Average Price).

58. The program storage device of claim 55, the method further comprising receiving one or more subscriptions from the one or more subscribers, the one or more subscriptions expressed as an SQL-like query.

59. The program storage device of claim 58, the method further comprising issuing the one or more subscriptions as continuous data stream queries which provide an asynchronous callback method.

60. The program storage device of claim 45, the method further comprising invoking a callback method provided by the particular one or more subscribers when the one or more publishers receives the portion of the filtered market that was filtered by the filter established by the particular one or more subscribers.

61. The program storage device of claim 59 wherein the asynchronous callback method is implemented using one or more of Linux Asynchronous 110 (A10), .NET Delegate and J2EE Message-Driven Beans (MDB).

62. The program storage device of claim 59, the method further comprising maximizing data parallelism using one or more of C/C++ using Asynchronous I/O (AI0), .NET Delegate, and J2EE.

63. The program storage device of claim 45 wherein a StreamBase™

stream-processing engine comprises the one or more queues and the one or more publishers.

64. The program storage device of claim 58, the method further comprising:

registering the one or more subscriptions using a remote procedure call (RPC) mechanism that includes a partition identifier; and

returning a subscription ID for subsequent use in subscription management.

65. The program storage device of claim 45, the method further comprising allocating costs to one or more consuming applications associated with the one or more subscribers, the allocation of costs being based at least in part on queries submitted by the one or more subscribers, and on the underlying resources consumed by the queries.

72. The program storage device of claim 45, wherein storing each of the extracted market data segments in a particular one of the one or more data partitions based on a characteristic of each of the extracted market data segments comprises storing each of the extracted market data segments in a particular one of the one or more data partitions based on an exchange symbol in the market data segment.

73. The program storage device of claim 45, wherein the one or more filters comprise at least a time predicate.

66. A method for low latency market data, the method comprising:

aggregating, by a first entity, one or more streams of market data, the market data comprising a plurality of market data segments;

storing, by the second entity, each of the market data segments in a particular one of one or more data partitions based on a characteristic of each of the extracted market data segments, the market data segments received via a direct connection, the direct connection comprising a cluster interconnect of Beowulf style;

filtering, by the second entity, the plurality of market data segments stored in the one or more data partitions using one or more filters established by a third entity, the filtering performed by creating a plurality of jobs and scheduling the plurality of jobs to be executed in parallel on one or more computer systems in a computational cluster;

registering, by a second entity, one or more queries regarding the one or more data streams;

publishing, by the second entity, one or more results for the one or more queries; and

subscribing, by the third entity, for receipt of at least one of the one or more results, wherein publishing one or more results for the one or more queries comprises;

determining whether the filtered data is associated with a shared subscription;

publishing the filtered data using an IP multicast channel if it is determined that the filtered data is associated with a shared subscription; and

publishing the filtered data using a point-to-point messaging transport if it is determined that the filtered data is not associated with a shared subscription; and

sending, via a callback mechanism, a notification to the third entity when the filtered data that was filtered by the filter established by the third entity is received.

67. An apparatus for low latency market data, the apparatus comprising:

a feed handler adapted to aggregate one or more streams of market data, the market data comprising a plurality of market data segments;

a message board coupled to the feed handler via a direct connection, the direct connection comprising a cluster interconnect of Beowulf style, the message board configured to;

store each of the market data segments in a particular one of one or more data partitions based on a characteristic of each of the extracted market data segments;

filter the plurality of market data segments stored in the one or more data partitions using one or more filters established by one or more subscribers, the filtering performed by creating a plurality of jobs and scheduling the plurality of jobs to be executed in parallel on one or more computer systems in a computational cluster;

register one or more queries regarding the one or more data streams;

publish one or more results for the one or more queries; and

send, via a callback mechanism, a notification to the one or more subscribers when the message board receives market data segments that were filtered by the one or more filters established by the one or more subscribers; and

one or more consumer applications configured to subscribe for receipt of at least one of the one or more results, wherein publishing one or more results for the one or more queries comprises;

determining whether the filtered data is associated with a shared subscription;

publishing the filtered data using an IP multicast channel if it is determined that the filtered data is associated with a shared subscription; and

publishing the filtered data using a point-to-point messaging transport if it is determined that the filtered data is not associated with a shared subscription.