Temporal affinity-based routing of workloads

• US 8,391,295 B2
• Filed: 10/05/2007
• Issued: 03/05/2013
• Est. Priority Date: 07/31/2007
• Status: Active Grant

• Current Assignee: Oracle International Corporation (Oracle Corporation)
• Original Assignee: Oracle International Corporation (Oracle Corporation)
• Inventors: Colrain, Carol, Zoll, Michael, Irudayaraj, Rajkumar
• Primary Examiner(s): Rinehart, Mark
• Assistant Examiner(s): ONAMUTI, GBEMILEKE J
• Application Number:
  US11/868,317
• Publication Number:
  US 20090034537A1
• Time in Patent Office: 1,978 Days
• Field of Search: 370/235, 370/400, 709/203, 709/229
• US Class Current: 370/400
• CPC Class Codes: H04L 45/121 by minimising delays
• CPC Class Codes: H04L 45/306 Route determination based on the nature of the carried application
• CPC Class Codes: H04L 67/1001 for accessing one among a plurality of replicated servers
• CPC Class Codes: H04L 67/1008 based on parameters of servers, e.g. available memory or workload monitoring of computer activity G06F11/30
• CPC Class Codes: H04L 67/61 taking into account QoS or priority requirements
• CPC Class Codes: H04L 67/63 Routing a service request depending on the request content or context

Abstract

Techniques for routing client requests among a group of nodes offering a service are described. A coordinator determines that performance could be improved by using affinity-based routing. In one embodiment, the coordinator calculates a Time-to-Live (TTL) metric whereby clients benefit by returning to the same cache and posts this hint to subscribers. Client'"'"'s start preserving locality data in an affinity context, such that later requests for a connection can be routed to the location last visited. The coordinator measures the system over subsequent intervals. If the gradient of the goodness (viz service quality and capacity) is stable or improving, then the coordinator continues to advise subscribers to use the affinity-based technique. Alternatively, if the gradient of the goodness is deteriorating, then the coordinator posts advice to the subscribers to stop using the affinity-based technique for any instance during the next intervals.

Claims

1. A computer-implemented method for distributing requests, the method comprising:

a coordinator within a database server cluster (a) developing advice as to whether affinity-based routing would benefit a service at a node in the database server cluster and (b) posting the advice to one or more middle-tier servers that send requests to and receive responses from database servers in the database server cluster;

determining whether a potential gain in performance in the database server cluster can be achieved by using affinity-based routing in the database server cluster;

in response to determining that a potential gain in performance in the database server cluster can be achieved by using affinity-based routing in the database server cluster, recommending, in the advice, to distribute requests to nodes in the database server cluster to preserve client-to-node affinity for requests from the same client;

after recommending, collating performance characteristics of the database server cluster;

based on collating the performance characteristics, measuring a potential loss in performance in the database server cluster;

in response to measuring a potential loss in performance in the database server cluster, removing the advice to distribute requests to nodes in the database server cluster to preserve client-to-node-affinity; and

wherein the method is performed by one or more computing devices.

2. The method of claim 1, further comprising:

determining an amount of time that affinity is expected to be beneficial based on a rate that data ages from caches of nodes in the database server cluster; and

based on determining an amount of time that affinity is expected to be beneficial,

publishing a time-to-live for affinity to be used by clients;

wherein affinity is temporal.

3. The method of claim 1, wherein the step of developing advice comprises developing advice based on a characteristic of the database server cluster, and wherein the characteristic is a weighting of change in goodness, including response time and capacity, across multiple iterations, and wherein goodness is computed using a formula:

((G(n)_max−

G(n)_min)−

(G(n−

1)_max−

G(n−

1)_min))/(G(n−

1)_max−

G(n−

1)_min) for each iteration (n), where G(n)_max is the maximum goodness measured during iteration (n), G(n)_min is the minimum goodness measured during iteration (n), G(n−

1)_maxis the maximum goodness measured during iteration (n−

1), and G(n−

1)_min is the minimum goodness measured during iteration (n−

1).

4. A computer-implemented method comprising:

obtaining a measurement of a characteristic of a system;

determining whether the measurement satisfies specified criteria; and

in response to determining that the measurement does not satisfy the specified criteria, distributing requests from one or more clients among two or more nodes of the system based on an affinity-based technique;

wherein distributing requests based on the affinity-based technique comprises sending all requests that come from each client of the one or more clients to a particular node of the two or more nodes while an affinity relationship exists between said each client and the particular node and a particular service; and

wherein the characteristic is an average time taken by nodes in the system to receive replicas of data blocks that were requested from other nodes in the system;

wherein each node in the system is a separate database server instance with a cache that stores data blocks;

wherein the method is performed by one or more computing devices.

5. The method of claim 4, wherein determining whether the measurement satisfies specified criteria comprises determining whether a gradient of goodness has deteriorated below a specified threshold.

6. The method of claim 4, wherein the characteristic is based on both (a) cluster wait time and (b) capacities of nodes in the system.

7. The method of claim 4, further comprising:

during a particular time interval, obtaining a measurement of a quality of the system;

determining whether a quality of the system has deteriorated to at least a specified degree during the particular time interval;

in response to determining that the quality has deteriorated to at least the specified degree during the particular time interval, posting advice to distribute, based on a load-balancing technique instead of an affinity-based technique, requests during a successive time interval that is subsequent to the particular time interval.

8. The method of claim 4, further comprising:

during a first time interval, obtaining a first measurement of a first cluster wait time in the system;

during a second time interval, obtaining a second measurement of a second cluster wait time in the system;

determining whether the second measurement is less than the first measurement; and

in response to determining that the second measurement is not less than the first measurement, posting advice to distribute, based on the load-balancing technique instead of the affinity-based technique, requests during a third time interval that is subsequent to the second time interval.

9. The method of claim 4, further comprising:

during a first time interval, obtaining a first measurement of a first cluster wait time in the system and determining that a potential advantage can be gained by enabling affinity;

during a second time interval, obtaining iterative measures of a goodness of the system;

determining whether a gradient of the goodness is improving or stable;

in response to determining that (a) the gradient of the goodness is stable or improving, providing advice to distribute, based on the affinity-based technique, requests for subsequent measurement intervals; and

iteratively verifying that the gradient of the goodness is stable or improving and updating the advice based on the gradient of the goodness.

10. The method of claim 4, further comprising:

establishing an affinity context that extends preservation of a location that the client last used to the client whereby the affinity context comprises sufficient data to route the client back to the same database and instance that was used on a last request for a connection to the server;

wherein the affinity context is available for later requests to provide connections at the same or different connection pool managers and listeners.

11. A computer-implemented method for distributing requests, the method comprising:

during a first time interval, distributing, based on a non-affinity load-balancing technique, requests from one or more clients among two or more nodes in a system;

during the first time interval, obtaining a measurement of a characteristic of the system;

determining whether the measurement satisfies specified criteria; and

if the measurement does not satisfy the specified criteria, then, during a second time interval that is subsequent to the first time interval, distributing requests from at least one of the one or more clients among the two or more nodes based on an affinity-based technique, thereby switching from the non-affinity load-balancing technique to the affinity-based technique in between the first time interval and the second time interval;

wherein distributing requests based on the affinity-based technique comprises sending all requests that come from a particular client to a particular node of the two or more nodes while an affinity relationship exists between the particular client and the particular node;

wherein the method is performed by one or more computing devices.

12. The method of claim 11, further comprising:

dissolving the affinity relationship between the particular client and the particular node in response to an expiration of a time-to-live period that applies to the affinity relationship.

13. The method of claim 11, further comprising:

if the measurement satisfies the specified criteria, then, during the second time interval, distributing, based on the load-balancing technique and not based on the affinity-based technique, requests from the one or more clients among the two or more nodes.

14. The method of claim 11, further comprising:

during the second time interval, obtaining a measurement of a quality of the system;

determining whether the quality has deteriorated to at least a specified degree during the second time interval;

in response to determining that the quality has deteriorated to at least the specified degree during the second time interval, distributing, based on the non-affinity load-balancing technique instead of the affinity-based technique, requests during a third time interval that is subsequent to the second time interval.

15. The method of claim 11, further comprising:

during the first time interval, obtaining a first measurement of a first cluster wait time in the system;

during the second time interval, obtaining a second measurement of a gradient of goodness for the system;

determining whether the second measurement is less than the first measurement; and

in response to determining that the second measurement is not less than the first measurement, distributing, based on the non-affinity load-balancing technique instead of the affinity-based technique, requests during a third time interval that is subsequent to the second time interval.

16. The method of claim 11, further comprising:

during the first time interval, obtaining a first measurement of a first cluster wait time in the system;

during the second time interval, obtaining a second measurement of a gradient of goodness of the system;

during the second time interval, obtaining a measurement of a quality of the system;

determining whether the second measurement is less than the first measurement;

determining whether the quality has deteriorated to at least a specified degree during the second time interval; and

in response to determining both that (a) the second measurement is less than the first measurement, and (b) the quality has not deteriorated to at least the specified degree during the second time interval, distributing, based on the affinity-based technique, requests during a third time interval that is subsequent to the second time interval.

17. A computer-readable non-transitory storage that stores instructions which, when executed by one or more processors, cause the one of more processors to perform the steps of:

a coordinator within a database server cluster (a) developing advice as to whether affinity-based routing would benefit a service at a node in the database server cluster and (b) posting the advice to one or more middle-tier servers that send requests to and receive responses from database servers in the database server cluster;

determining whether a potential gain in performance in the database server cluster can be achieved by using affinity-based routing in the database server cluster;

in response to determining that a potential gain in performance in the database server cluster can be achieved by using affinity-based routing in the database server cluster, recommending, in the advice, to distribute requests to nodes in the database server cluster to preserve client-to-node affinity for requests from the same client;

after recommending, collating performance characteristics of the database server cluster;

based on collating the performance characteristics, measuring a potential loss in performance in the database server cluster; and

in response to measuring a potential loss in performance in the database server cluster, removing the advice to distribute requests to nodes in the database server cluster to preserve client-to-node affinity.

18. The computer-readable non-transitory storage of claim 17, further comprising:

determining an amount of time that affmity is expected to be beneficial based on a rate that data ages from caches of nodes in the database server cluster; and

based on determining an amount of time that affmity is expected to be beneficial, publishing a time-to-live for affmity to be used by clients;

wherein affmity is temporal.

19. The computer-readable non-transitory storage of claim 17, wherein the step of developing advice comprises developing advice based on a characteristic of the database server cluster, and wherein the characteristic is a weighting of change in goodness, including response time and capacity, across multiple iterations, and wherein goodness is computed using a formula:

((G(n)_max—

G(n)_min)—

(G(n−

1)_max—

G(n−

1)_min))/(G(n−

1)<sub>max—

G(n−</sub>

1)_min) for each iteration (n), where G(n)_max is the maximum goodness measured during iternation (n), G(n)_min is the minimum goodness measured during iteration (n), G(n−

1)_max is the maximum goodness measured during iteration (n−

1), and G(n−

1)_min is the minimum goodness measured during iteration (n−

1).

20. A computer-readable non-transitory storage that stores instructions which, when executed by one or more processors, cause the one of more processors to perform the steps of:

obtaining a measurement of a characteristic of a system;

determining whether the measurement satisfies specified criteria; and

in response to determining that the measurement does not satisfy the specified criteria, distributing requests from one or more clients among two or more nodes of the system based on an affinity-based technique;

wherein distributing requests based on the affinity-based technique comprises sending all requests that come from each client of the one or more clients to a particular node of the two or more nodes while an affinity relationship exists between said each client and the particular node and a particular service; and

wherein the characteristic is an average time taken by nodes in the system to receive replicas of data blocks that were requested from other nodes in the system; and

wherein each node in the system is a separate database server instance with a cache that stores data blocks.

21. The computer-readable non-transitory storage of claim 20, wherein determining whether the measurement satisfies specified criteria comprises determining whether a gradient of goodness has deteriorated below a specified threshold.

22. The computer-readable non-transitory storage of claim 20, wherein the characteristic is based on both (a) cluster wait time and (b) capacities of nodes in the system.

23. The computer-readable non-transitory storage of claim 20, further comprising:

during a particular time interval, obtaining a measurement of a quality of the system;

determining whether a quality of the system has deteriorated to at least a specified degree during the particular time interval;

in response to determining that the quality has deteriorated to at least the specified degree during the particular time interval, posting advice to distribute, based on a load-balancing technique instead of an affmity-based technique, requests during a successive time interval that is subsequent to the particular time interval.

24. The computer-readable non-transitory storage of claim 20, further comprising:

during a first time interval, obtaining a first measurement of a first cluster wait time in the system;

during a second time interval, obtaining a second measurement of a second cluster wait time in the system;

determining whether the second measurement is less than the first measurement; and

in response to determining that the second measurement is not less than the first measurement, posting advice to distribute, based on the load-balancing technique instead of the affmity-based technique, requests during a third time interval that is subsequent to the second time interval.

25. The computer-readable non-transitory storage of claim 20, further comprising:

during a first time interval, obtaining a first measurement of a first cluster wait time in the system and determining that a potential advantage can be gained by enabling affinity;

during a second time interval, obtaining iterative measures of a goodness of the system;

determining whether a gradient of the goodness is improving or stable;

in response to determining that (a) the gradient of the goodness is stable or improving, providing advice to distribute, based on the affmity-based technique, requests for subsequent measurement intervals; and

iteratively verifying that the gradient of the goodness is stable or improving and updating the advice based on the gradient of the goodness.

26. The computer-readable non-transitory storage of claim 20, further comprising:

establishing an affmity context that extends preservation of a location that the client last used to the client whereby the affmity context comprises sufficient data to route the client back to the same database and instance that was used on a last request for a connection to the server;

wherein the affmity context is available for later requests to provide connections at the same or different connection pool managers and listeners.

27. A computer-readable non-transitory storage that stores instructions which, when executed by one or more processors, cause the one of more processors to perform the steps of:

during a first time interval, distributing, based on a non-affinity load-balancing technique, requests from one or more clients among two or more nodes in a system;

during the first time interval, obtaining a measurement of a characteristic of the system;

determining whether the measurement satisfies specified criteria; and

if the measurement does not satisfy the specified criteria, then, during a second time interval that is subsequent to the first time interval, distributing requests from at least one of the one or more clients among the two or more nodes based on an affinity-based technique, thereby switching from the non-affinity load-balancing technique to the affinity-based technique in between the first time interval and the second time interval; and

wherein distributing requests based on the affinity-based technique comprises sending all requests that come from a particular client to a particular node of the two or more nodes while an affinity relationship exists between the particular client and the particular node.

28. The computer-readable non-transitory storage of claim 27, further comprising:

dissolving the affmity relationship between the particular client and the particular node in response to an expiration of a time-to-live period that applies to the affinity relationship.

29. The computer-readable non-transitory storage of claim 27, further comprising:

if the measurement satisfies the specified criteria, then, during the second time interval, distributing, based on the load-balancing technique and not based on the affmity- based technique, requests from the one or more clients among the two or more nodes.

30. The computer-readable non-transitory storage of claim 27, further comprising:

during the second time interval, obtaining a measurement of a quality of the system;

determining whether the quality has deteriorated to at least a specified degree during the second time interval;

in response to determining that the quality has deteriorated to at least the specified degree during the second time interval, distributing, based on the non-affinity load-balancing technique instead of the affmity-based technique, requests during a third time interval that is subsequent to the second time interval.

31. The computer-readable non-transitory storage of claim 27, further comprising:

during the first time interval, obtaining a first measurement of a first cluster wait time in the system;

during the second time interval, obtaining a second measurement of a gradient of goodness for the system;

determining whether the second measurement is less than the first measurement; and

in response to determining that the second measurement is not less than the first measurement, distributing, based on the non-affinity load-balancing technique instead of the affmity-based technique, requests during a third time interval that is subsequent to the second time interval.

32. The computer-readable non-transitory storage of claim 27, further comprising:

during the first time interval, obtaining a first measurement of a first cluster wait time in the system;

during the second time interval, obtaining a second measurement of a gradient of goodness of the system;

during the second time interval, obtaining a measurement of a quality of the system;

determining whether the second measurement is less than the first measurement;

determining whether the quality has deteriorated to at least a specified degree during the second time interval; and

in response to determining both that (a) the second measurement is less than the first measurement, and (b) the quality has not deteriorated to at least the specified degree during the second time interval, distributing, based on the affinity-based technique, requests during a third time interval that is subsequent to the second time interval.