SYSTEM AND METHOD FOR OPERATING A SERVER FARM

• US 20170262317A1
• Filed: 03/11/2016
• Published: 09/14/2017
• Est. Priority Date: 03/11/2016
• Status: Active Application

• Current Assignee: City University of Hong Kong
• Original Assignee: City University of Hong Kong
• Inventors: Wong, Wing Ming Eric, Fu, Jing, Guo, Jun, Zukerman, Moshe
• Granted Patent: US 10,268,523 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 9/5094 where the allocation takes into account power or heat criteria power management in computers in general G06F1/3203; thermal management in computers in general G06F1/206
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

A method for operating a server farm with a plurality of servers operably connected with each other includes the steps of: sorting the plurality of servers according to a respective energy efficiency value associated with each of the plurality of servers; defining a virtual server by selectively joining two or more of the servers with the highest energy efficiency values; receiving a job request of a computational task to be handled by the server farm; and assigning the computational task to one of the plurality of servers for processing based on a job assignment policy, the job assignment policy being arranged to give preference to assigning the computational task to the servers of the virtual server for processing so as to maximize an energy efficiency of the server farm.

Claims

1. A method for operating a server farm with a plurality of servers operably connected with each other, comprising the steps of:

sorting the plurality of servers according to a respective energy efficiency value associated with each of the plurality of servers;

defining a virtual server by selectively joining two or more of the servers with the highest energy efficiency values;

receiving a job request of a computational task to be handled by the server farm; and

assigning the computational task to one of the plurality of servers for processing based on a job assignment policy, the job assignment policy being arranged to give preference to assigning the computational task to the servers of the virtual server for processing so as to maximize an energy efficiency of the server farm.

2. The method in accordance with claim 1, further comprising the steps of:

monitoring a number or a rate of job request of computational task received; and

updating a definition of the virtual server by adjusting the number of servers defining the virtual server based on the monitoring result.

3. The method in accordance with claim 1, wherein the plurality of servers each includes a finite buffer for queuing job requests, and the server farm is finite-buffer processor-sharing server farm.

4. The method in accordance with claim 1, wherein the job assignment policy is arranged to give preference to assigning the computational task to one of the servers of the virtual server with at least one vacant slot in its buffer.

5. The method in accordance with claim 1, wherein the job assignment policy is arranged to reassign a computational task being handled by one of the plurality of servers to another one of the plurality of servers that is more energy efficient, before the computational task is completed.

6. The method in accordance with claim 1, wherein the job assignment policy is arranged to reassign a computational task being handled by a server in the virtual server to another server in the virtual server with buffer vacancy before the computational task is completed.

7. The method in accordance with claim 1, wherein the job assignment policy is arranged to ensure that none of the servers of the virtual server is idle if a server with a lower energy efficiency value in the virtual server is busy.

8. The method in accordance with claim 1, wherein the job assignment policy is arranged to ensure that none of the servers of the virtual server handles more than one computational task if any of the servers of the virtual server is idle.

9. The method in accordance with claim 1, wherein the job assignment policy is arranged to ensure that none of the servers other than those of the virtual server is busy if a server with a higher energy efficiency value has at least one vacant slot in its buffer.

10. The method in accordance with claim 1, wherein the virtual server is formed by at least two servers with the highest efficiency values in the server farm.

11. The method in accordance with claim 1, wherein the number of servers of the virtual server is selected such that an aggregate service rate of the virtual server is less than or equal to an arrival rate of the job requests;

the aggregate service rate of the virtual server being defined as a sum of service rate of each server in the virtual server.

12. The method in accordance with claim 1, wherein the number of servers of the virtual server is selected such that an aggregate service rate of the virtual server substantially matches an average arrival rate of the job requests;

the aggregate service rate of the virtual server being defined as a sum of service rate of each server in the virtual server.

13. The method in accordance with claim 1, wherein the energy efficiency value of a server is defined by a service rate of the server divided by an energy consumption rate of the server.

14. A system for operating a server farm with a plurality of servers operably connected with each other, the system comprising one or more processors arranged to:

sort the plurality of servers according to a respective energy efficiency value associated with each of the plurality of servers;

define a virtual server by selectively joining two or more of the servers with the highest energy efficiency values;

receiving a job request of a computational task to be handled by the server farm; and

assign the computational task to one of the plurality of servers for processing based on a job assignment policy programmed to be operated in the one or more processors, the job assignment policy being arranged to give preference to assigning the computational task to the servers of the virtual server for processing so as to maximize an energy efficiency of the server farm.

15. The system in accordance with claim 14, wherein the one or more processors are further arranged to:

monitor a number or a rate of job request of computational task received; and

update a definition of the virtual server by adjusting the number of servers defining the virtual server based on the monitoring result.

16. The system in accordance with claim 14, wherein the plurality of servers each includes a finite buffer for queuing job requests, and the server farm is finite-buffer processor-sharing server farm.

17. The system in accordance with claim 14, wherein the job assignment policy is arranged to give preference to assigning the computational task to one of the servers of the virtual server with at least one vacant slot in its buffer.

18. The system in accordance with claim 14, wherein the job assignment policy is arranged to reassign a computational task being handled by one of the plurality of servers to another one of the plurality of servers that is more energy efficient, before the computational task is completed.

19. The system in accordance with claim 14, wherein the job assignment policy is arranged to reassign a computational task being handled by a server in the virtual server to another server in the virtual server with buffer vacancy before the computational task is completed.

20. The system in accordance with claim 14, wherein the job assignment policy is arranged to ensure that none of the servers of the virtual server is idle if a server with a lower energy efficiency value in the virtual server is busy.

21. The system in accordance with claim 14, wherein the job assignment policy is arranged to ensure that none of the servers of the virtual server handles more than one computational task if any of the servers of the virtual server is idle.

22. The system in accordance with claim 14, wherein the job assignment policy is arranged to ensure that none of the servers other than those of the virtual server is busy if a server with a higher energy efficiency value has at least one vacant slot in its buffer.

23. The system in accordance with claim 14, wherein the virtual server is formed by at least two servers with the highest efficiency values in the server farm.

24. The system in accordance with claim 14, wherein the number of servers of the virtual server is selected such that an aggregate service rate of the virtual server is less than or equal to an arrival rate of the job requests;

the aggregate service rate of the virtual server being defined as a sum of service rate of each server in the virtual server.

25. The system in accordance with claim 14, wherein the number of servers of the virtual server is selected such that an aggregate service rate of the virtual server substantially matches an average arrival rate of the job requests;

the aggregate service rate of the virtual server being defined as a sum of service rate of each server in the virtual server.

26. The system in accordance with claim 14, wherein the energy efficiency value of a server is defined by a service rate of the server divided by an energy consumption rate of the server.

27. The system in accordance with claim 14, wherein the one or more processors are one or more servers in the server farm.

28. A non-transient computer readable medium for storing computer instructions that, when executed by one or more processors, causes the one or more processors to perform a method for operating a server farm with a plurality of servers operably connected with each other, the method comprising the steps of:

sorting the plurality of servers according to a respective energy efficiency value associated with each of the plurality of servers;

defining a virtual server by selectively joining two or more of the servers with the highest energy efficiency values;

receiving a job request of a computational task to be handled by the server farm; and

assigning the computational task to one of the plurality of servers for processing based on a job assignment policy, the job assignment policy being arranged to give preference to assigning the computational task to the servers of the virtual server for processing so as to maximize an energy efficiency of the server farm.