Mechanism for controlled server overallocation in a datacenter

Patent metadata

Field	Value
Publication number	US-10198295-B2
Application number	US-201515310020-A
Country	US
Kind code	B2
Filing date	May 19, 2015
Priority date	May 21, 2014
Publication date	Feb 5, 2019
Grant date	Feb 5, 2019

Abstract

A method of controlling a datacenter ( 1 ), for example a cloud datacenter, and a computer management system ( 3 ) for managing a datacenter comprising a plurality of servers ( 2 ) configured to execute a plurality of jobs is provided. The method comprises receiving a request to execute a job from a user ( 4 ), determining an allocated server ( 2 ) on which to execute the job, and executing the job on the allocated server ( 2 ). The determining the allocated server ( 2 ) on which to execute the job comprises: classifying the job according to its resource requirement, selecting a subset of the servers ( 2 ) that fulfill the resource requirements of the job, determining the allocated server ( 2 ) that can execute the job with a favorable energy efficiency, wherein a total resource estimate for all jobs running on the server ( 2 ) from time to time exceeds the resources of at least one of the servers ( 2 ). Accordingly, the datacenter ( 1 ) may over allocate the capacity of the physical servers ( 2 ), to overcome the overestimation of required resources by users ( 4 ).

First claim

The invention claimed is:

1. A method of controlling a datacentre comprising a plurality of servers to execute a plurality of jobs, the method comprising:
receiving a request to execute a job from a user;
determining an allocated server on which to execute the job; and
executing the job on the allocated server,
wherein the determining comprises:
(i) classifying the job according to its resource requirement;
(ii) selecting a subset of the servers that fulfil the resource requirements of the job; and
(iii) from the selected subset, determining as the allocated server a server that can execute the job with a favourable energy efficiency, said determining the server from the selected subset that can execute the job with a favourable energy efficiency includes:
calculating an overallocation rate, OAR, of each server in the subset of servers;
determining if each server in the subset has available resources to execute the job given the calculated OAR;
determining the estimated energy efficiency of each server in the subset if the job is assigned to the server; and
assigning the server in the subset which has available resources and a highest estimated energy efficiency to be the allocated server; and

wherein a total resource estimate for all jobs running on a server in the plurality from time to time exceeds the resources of at least one of the servers in the plurality.

2. The method of claim 1 , wherein the calculating the OAR comprises:
calculating an overselling profit rate (P) of the server;
establishing a category of the user;
calculating the OAR from a statistical distribution describing overestimation patterns of the category of the user, using the overselling profit rate (P).

3. The method of claim 2 , wherein the overselling profit rate is calculated according to the formula:

P
=

R

R
+
C

wherein R is a difference in energy efficiency (EE) caused by allocating the job to the server, and
wherein C is the EE after allocation multiplied by the estimated change in EE.

4. The method of claim 3 , wherein the estimated change in EE is calculated using an estimated performance interference.

5. The method of claim 4 , wherein the estimated performance interference is an estimated aggregate of a performance interference produced by the plurality of jobs co-located on the server, based on a class of each job located on the server.

6. The method of claim 1 , wherein the classifying the job comprises:
training a decision tree classifier on a training set of historical tracelog data;
classifying the job with the decision tree classifier.

7. The method of claim 1 , wherein the selecting the subset of the servers comprises:
calculating a similarity level between a set of server features (F) and a set of job constraints (C), and
selecting the servers having the similarity level equal to or greater than a predetermined minimum similarity value.

8. The method of claim 1 , wherein the method further comprises:
detecting an overload event in at least one of the plurality of servers;
resolving the detected overload event by evicting, pausing or migrating jobs having a lowest priority and a shortest running time.

9. A computer management system, CMS, for managing a computer datacentre comprising a plurality of servers, each comprising at least one virtual container, VC, the CMS comprising:
one or more processors; and
memory with computer code instructions stored thereon, the one or more processors and the memory with the computer code instructions being configured to cause the CMS to:
receive a request to execute a job from a user, to control the CMS to determine a server on which the job will be executed and to control the CMS to execute the job;
classify the job according to its resource requirements;
select a subset of servers which fulfil a set of constraints of the job;
record the status of each server in the datacentre;
communicate with the plurality of servers, and
determine the server which can execute the job with a favourable energy efficiency from the subset of servers and a class of the job; calculate an overallocation rate, OAR, of each server in the subset of servers; determine if each server in the subset has available resources to execute the job given the OAR; and determine the estimated energy efficiency of each server in the subset if the job is assigned to the server.

10. The system of claim 9 , wherein the one or more processors and the memory with the computer code instructions are further configured to cause the system to detect an overload event in at least one of the plurality of servers and resolve the detected overload event by evicting, pausing or migrating jobs having a lowest priority and a shortest running time.

11. The system of claim 9 , wherein the one or more processors and the memory with the computer code instructions are further configured to cause the system to:
establish a category of a user and determine a statistical distribution describing overestimation patterns of the category of the user.

12. The system of claim 11 , wherein the one or more processors and the memory with the computer code instructions are further configured to cause the system to:
calculate an overselling profit rate (P) of the server; and
calculate the OAR from the statistical distribution describing overestimation patterns of the category of the user, using the overselling profit rate (P).

13. The system of claim 12 , wherein the one or more processors and the memory with the computer code instructions are further configured to cause the system to calculate the overselling profit rate according to the formula:

P
=

R

R
+
C

wherein R is a difference in energy efficiency (EE) caused by allocating the job to the server, and
wherein C is the EE after allocation multiplied by the estimated change in EE.

14. The system of claim 9 , wherein, in selecting a subset of servers which fulfil a set of constraints of the job, the one or more processors and the memory with the computer code instructions are further configured to cause the system to:
calculate a similarity level between a set of server features (F) and a set of job constraints (C), and
select the servers having the similarity level equal to or greater than a predetermined minimum similarity value.

15. A computer datacentre comprising:
a plurality of servers, each comprising at least one virtual container, VC, wherein the computer datacentre incorporates a computer management system, CMS, as claimed in claim 9 .

16. A non-transitory computer-readable storage medium comprising:
computer-executable components which when executed cause a computing device to perform the method of claim 1 .

Assignees

• Univ Leeds Innovations Ltd

Inventors

• Moreno Ismael Solis
• Townend Paul Michael
• Garraghan Peter Michael
• Xu Jie

Classifications

• Y02D10/22

  Cross-Sectional Technologies · mapped topic

• G06F9/4893

  taking into account power or heat criteria (power management in computers in general G06F1/3203; thermal management in computers in general G06F1/206) · CPC title

• G06F9/5044 Primary

  considering hardware capabilities · CPC title

• H04L47/70

  Admission control; Resource allocation · CPC title

• G06F9/5094

  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 title