Self-evolving distributed system performance using a system health index

• US 7,698,239 B2
• Filed: 04/28/2006
• Issued: 04/13/2010
• Est. Priority Date: 03/25/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Samuel, Geogy A., Norton, Noah, Lieuallen, Brian R., Singhal, Sandeep Kishan, Manion, Todd R.
• Primary Examiner(s): Vincent; David R.
• Assistant Examiner(s): CHANG, LI WU
• Application Number:
  US11/413,538
• Publication Number:
  US 20060262726A1
• Time in Patent Office: 1,446 Days
• Field of Search: 706/16
• US Class Current: 706/16
• CPC Class Codes: H04L 41/145 involving simulating, designing, planning or modelling of a network
• CPC Class Codes: H04L 41/5003 Managing SLA; Interaction between SLA and QoS
• CPC Class Codes: H04L 43/0817 by checking functioning
• CPC Class Codes: H04L 43/50 Testing arrangements
• CPC Class Codes: H04L 63/0272 Virtual private networks

Abstract

In a distributed computing system, an artificial intelligence system may be employed to configure the network variables. A metric describing the overall system performance may be derived during network operation or simulation and compared to an ideal metric describing the same distributed system performance. The difference between the derived metric and the ideal metric may then be used with an artificial intelligence system to modify the network variables to evolve the system toward the ideal performance standard.

Claims

1. A method of improving the performance of a distributed computing system comprising:

determining an ideal system health index that is a quantifiable representation of a user-defined performance characteristic of the distributed system, the distributed system comprising a peer-to-peer network including a plurality of nodes, the ideal system health index including a plurality of system performance metrics collected from the plurality of nodes, each metric modified by one or more system variables;

propagating one or more system performance metrics through an artificial intelligence system, the system performance metrics measured during operation of the distributed system, the propagating including;

multiplying each system performance metric by one or more first weights,

processing each weighted system performance metric by one or more functions within one or more hidden layers of the artificial intelligence system, each function accounting for the dependence of each system performance metric on one or more other system performance metrics, and

converting each weighted and processed system performance metric to a plurality of improved system variables by multiplying each weighted and processed system performance metric by one or more second weights;

modifying one or more system variables with the plurality of improved system variables;

determining a system health index by aggregating the system performance metrics after modifying one or more system variables with the plurality of improved system variables;

comparing the system health index to the ideal system health index to derive a health index error value; and

when the health index error value is above a threshold;

identifying a slow or failing node of the plurality of nodes in the peer-to-peer network based on the system health index,

correcting the plurality of improved distributed system variables of the slow or failing node by applying the health index error value to the plurality of improved distributed system variables of the slow or failing node;

if the health index error value is below the threshold, saving the improved distributed system variables,

modifying the artificial intelligence system by applying a learning algorithm to the artificial intelligence system to reduce the health index error value, the learning algorithm using the health index error value,

re-propagating the system performance metrics through the modified artificial intelligence system to derive a second plurality of improved system variables, and

applying the second plurality of improved system variables to at least the slow or failing node.

2. The method of claim 1, wherein applying the health index error value to the improved distributed system variables comprises configuring the artificial intelligence system based on a comparison of the system health index and the ideal system health index.

3. The method of claim 1, further comprising:

applying a learning algorithm to the artificial intelligence system after comparing the system health index to the ideal system health index to derive the health index error value; and

deriving at least one second improved system variable from the artificial intelligence system after applying the learning algorithm.

4. The method of claim 3 wherein the learning algorithm is a back propagation algorithm.

5. The method of claim 1, wherein the at least one improved system variable comprises a system timeout value, a distributed hash table keyspace size, a routing table size, a maximum number of distributed system nodes, or a maximum number of allowable packet hops.

6. The method of claim 1, wherein the first system health index and the ideal system health index comprises one or more of a number of distributed system nodes, an average distributed system latency, an average distributed system packet loss, an operation frequency, or a distributed system node cache churn rate.

7. The method of claim 1, wherein the distributed system is one of a live distributed system or a simulated distributed system.

8. The method of claim 1, wherein the at least one artificial intelligence system is a neural network.

9. The method of claim 8, wherein the neural network comprises at least one hidden layer node, the at least one hidden layer node comprising the one or more functions, each function representing an interaction between a plurality of system behavior variables.

10. The method of claim 1, wherein instrumentation at each node of the distributed system collects performance metrics for that node.

11. The method of claim 10, further comprising reporting node performance statistics to another node outside the distributed system to aggregate the statistics into the system health index.

12. A computing device comprising:

a processor;

a memory; and

an application maintained in the memory and executed on the processor to improve the performance of a distributed computing system, the distributed computing system comprising a peer-to-peer network including a plurality of nodes, the application comprising instructions for;

determining an ideal system health index that is a quantifiable representation of a user-defined performance characteristic of the distributed system, the ideal system health index including a plurality of system performance metrics collected from the plurality of nodes, each metric modified by one or more system variables;

propagating one or more system performance metrics through a neural network including an input layer, one or more hidden layers, and an output layer, the system performance metrics measured during operation of the distributed system, the propagating including;

multiplying each system performance metric by one or more first weights at the input layer,

processing each weighted system performance metric by one or more functions within one or more hidden layers of the neural network, each function accounting for the dependence of each system performance metric on one or more other system performance metrics, and

converting each weighted and processed system performance metric to a first plurality of improved system variables by multiplying each weighted and processed system performance metric by one or more second weights at the output layer;

modifying the system variables with the first plurality of improved system variables;

determining a system health index by aggregating the system performance metrics after modifying one or more system variables with the first plurality of improved system variables;

comparing the system health index to the ideal system health index to derive a health index error value; and

when the health index error value is above a threshold;

identifying a slow or failing node of the plurality of nodes in the peer-to-peer network based on the system health index,

correcting the first plurality of improved distributed system variables of the slow or failing node by applying the health index error value to the first plurality of improved distributed system variables of the slow or failing node,

modifying the neural network by applying a back propagation algorithm to the neural network, the back propagation algorithm using the health index error value to modify one or more values of the input layer, the hidden layer, and the output layer, and

re-propagating the system performance metrics through the modified neural network to derive a second plurality of improved system variables, and

applying the second plurality of improved system variables to at least the slow or failing node.

if the health index error value is below the threshold, saving the improved distributed system variables and operating the distributed system using the improved distributed system variables.

13. The computing device of claim 12, wherein the distributed system is one of a live distributed system or a simulated distributed system.

14. The method of claim 12, wherein instrumentation at each node of the distributed system collects performance metrics for that node.

15. The method of claim 14, further comprising reporting node performance statistics to another node outside the distributed system to aggregate the statistics into the system health index.

16. A computer storage medium comprising computer-executable instructions for improving the performance of a distributed computing system, the instructions comprising:

determining an ideal system health index that is a quantifiable representation of a user-defined performance characteristic of the distributed system, the distributed system comprising a peer-to-peer network including a plurality of nodes, the ideal system health index including a plurality of system performance metrics collected from the plurality of nodes, each metric modified by one or more system variables;

propagating one or more system performance metrics through a neural network including an input layer, one or more hidden layers, and an output layer, the system performance metrics measured during operation of the distributed system, the propagating including;

multiplying each system performance metric by one or more first weights at the input layer,

processing each weighted system performance metric by one or more functions within one or more hidden layers of the neural network, each function accounting for the dependence of each system performance metric on one or more other system performance metrics, and

converting each weighted and processed system performance metric to a plurality of improved system variables by multiplying each weighted and processed system performance metric by one or more second weights at the output layer;

modifying the system variables with the plurality of improved system variables;

determining a system health index by aggregating the system performance metrics after modifying one or more system variables with the plurality of improved system variables;

comparing the system health index to the ideal system health index to derive a health index error value; and

when the health index error value is above a threshold;

identifying a slow or failing node of the plurality of nodes in the peer-to-peer network based on the system health index,

correcting the plurality of improved distributed system variables by applying the health index error value to the plurality of improved distributed system variables of at least the slow or failing node,

modifying the neural network by applying a back propagation algorithm to the neural network, the back propagation algorithm using the health index error value to modify one or more values of the input layer, the hidden layer, and the output layer, and

re-propagating the system performance metrics through the modified neural network to derive a second plurality of improved system variables, and

applying the second plurality of improved system variables to at least the slow or failing node,

if the health index error value is below the threshold, saving the improved distributed system variables and operating the distributed system using the improved distributed system variables.

17. The computer storage medium of claim 16, wherein the computer executable instructions for applying the health index error value to the improved distributed system variables comprises configuring the artificial intelligence system based on the result of comparing the system health index and the ideal system health index.

18. The computer storage medium of claim 16, wherein the improved system variable comprises one or more of a system timeout value, a distributed hash table keyspace size, a routing table size, a maximum number of distributed system nodes, or a maximum number of allowable packet hops.

19. The computer storage medium of claim 16, wherein the distributed system is one of a live distributed system or a simulated distributed system.

20. The computer storage medium of claim 16, wherein the neural network comprises at least one hidden layer node, the at least one hidden layer node comprising a function, the function representing an interaction between multiple system variables.