Saving power by managing the state of inactive computing devices according to specific constraints

• US 8,904,213 B2
• Filed: 08/19/2013
• Issued: 12/02/2014
• Est. Priority Date: 11/04/2010
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Chan, Hoi Y., Das, Rajarshi, Hanson, James E., Isci, Canturk, Kephart, Jeffrey O., Levine, David W.
• Primary Examiner(s): Lee, Thomas
• Assistant Examiner(s): NGUYEN, PHIL K
• Application Number:
  US13/969,793
• Publication Number:
  US 20140059367A1
• Time in Patent Office: 470 Days
• Field of Search: 713/320, 713/324
• US Class Current: 713/320
• CPC Class Codes: G06F 1/3206 Monitoring of events, devices or parameters that trigger a change in power modality
• CPC Class Codes: G06F 1/3209 Monitoring remote activity, e.g. over telephone lines or network connections
• CPC Class Codes: G06F 1/3234 Power saving characterised by the action undertaken
• CPC Class Codes: G06F 1/3287 by switching off individual functional units in the computer system
• CPC Class Codes: G06F 1/329 by task scheduling
• 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: H04L 67/00 Network arrangements or protocols for supporting network services or applications user-to-user messaging H04L51/00; network arrangements, protocols or services for supporting real-time applications in data packet communications networks H04L65/00
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

A system method and computer program product for managing readiness states of a plurality of computing devices. In response to a request, a computer system operates to either: provide one or more computing devices from an inactive pool to an active pool, or accept one or more active computing devices into the inactive pool. An Inactive Pool Manager proactively manages the inactive states of each computing device by: determining the desired number (and identities) of computing devices to be placed in each inactive state of readiness by solving a constraint optimization problem that describes a user-specified trade-off between expected readiness (estimated time to be able to activate computing devices when they are needed next) and conserving energy; generating a plan for changing the current set of inactive states to the desired set; and, executing the plan. Multiple alternative ways of quantifying the desired responsiveness to surges in demand are provided.

Claims

1. A system for managing inactive states of computing devices comprising:

a memory,

a processor unit in communication with said memory, for performing a method comprising;

receiving, at said processor unit, a request to one of;

activate or deactivate one or more computing devices from a respective group of inactive or active computing devices, each said inactive or active computing device initially in a pre-determined state of readiness;

activating or deactivating said one or more computing devices, said activating or deactivating resulting in a number of computing devices in one or more activation states of readiness i;

computing an optimal configuration of said computing devices in said one or more said states of readiness i by minimizing an expected power consumption Σ

_i A_i P_i where i represents all inactive computing device states, A_i represents a number of said computing devices currently in each state i, and, P_i represents an average power consumption of computing devices while in said state i; and

placing said computing devices in said optimized configuration, said processor unit is further configured to;

formulate an objective function to be minimized, said objective function representing said expected power consumption Σ

_i A_iP_i subject to a constraint,

wherein, after one of said activating or deactivating one or more computing devices, said processor unit is further configured to;

compute activation vector A″

representing a number of computing devices in said optimized one or more activation states of readiness i;

specify an upper threshold T_max representing a maximum expected time to activate additional computing devices; and

compute probabilities p_i for i additional computing devices to be needed, where i ranges from 1 to a specified upper threshold, wherein said activation vector A″

is computed by formulating said objective function to minimize an expected power consumption Σ

_i A″

_i P_i given a constraint Σ

_iR_ip_i<

T_max, where Σ

_iR_ip_i is a weighted sum representing the expected time required to make the i computing devices available.

2. The system as claimed in claim 1, wherein, after one of said activating or deactivating one or more computing devices, said processor unit is further configured to:

compute a resulting activation vector A′

representing said resulting number of computing devices in one or more activation states of readiness i.

3. The system as claimed in claim 2, wherein said processor unit is further configured to:

initiate transition of said computing devices from one inactive state to another to reach a desired power state for each computing device, said processor unit generating a plan to minimize a time required to accomplish said transitioning from said activation vector A′

to optimized activation vector A″

according to a transition latency matrix T_ij, representing an expected time required to change an activation state of a computing device from activation state i to activation state j.

4. The system as claimed in claim 2, wherein said objective function is subject to a constraint that R_k≦

R⁰_k for each computing devices k, wherein said R represents a readiness vector, a k^th component R_k representing a minimal amount of time for k computing devices to be restored to an active state given a specific activation vector A and said transition latency matrix T; and

, R⁰ represents a specified readiness vector, a k^th component R⁰_k representing a specified constraint on a maximum time permitted for k inactive servers to be restored to said active state.

5. The system as claimed in claim 2, wherein said processor unit is further configured to:

specify a desired demand fluctuation response vector F⁰ such that a k^th component F⁰_k represents a maximum time permitted for a plurality of computing devices to be transitioned in anticipation of a workload intensity increase above a current level; and

convert a requirement expressed in terms of said fluctuation response vector F⁰ to said readiness vector R⁰.

6. The system as claimed in claim 2, further comprising:

computing said probabilities p_s for “

s”

computing devices by;

specifying a c(w) that represents a cumulative workload intensity distribution as a function of a workload intensity “

w”

;

computing a maximal workload intensity level w(s) for which “

s”

servers can attain a pre-determined performance level w(s);

computing a corresponding cumulative distribution value c(w(s)) to denote a probability that an observed cluster-level workload will be less than or equal to w(s); and

computing probability p_s for “

s”

servers to be required according to;

p_s=c(w(s))−

c(w(s−

1))

where cumulative density values c(w(s)) and c(w(s−

1)) correspond respectively to attainable workload intensity for an amount s servers w(s) and the attainable workload intensity for an amount s−

1 servers.

7. The system as claimed in claim 2, wherein, after calculating said desired activation vector A″

, said processor unit is further configured to;

modify the desired activation vector into a user-preferred Activation Vector A′

″

; and

input, to a planner device, a sequence of actions for transitioning the computing devices according to said user-preferred Activation Vector A′

″

.

8. The system as claimed in claim 2, wherein said processor unit is further configured to:

receive, at said processor unit, a request to de-activate one or more active computing devices; and

placing said de-activated computing devices at a specified state of readiness, thereby, resulting in a new activation vector A′

;

solve a constraint optimization problem for computing a best activation vector A″

from said activation state vector A′

;

generate a plan that minimizes a sequence of actions required to switch state of said computing devices from activation state A′

to activation state A″

; and

execute the plan to bring said computing devices to said activation state A″

.

9. A computer program product for managing inactive states of computing devices, the computer program product comprising:

a storage medium that is not a propagating signal readable by a processing circuit and storing instructions for execution by the processing circuit for performing a method comprising;

receiving a request to one of;

activate or deactivate one or more computing devices from a respective group of inactive or active computing devices, each said inactive or active computing device initially in a pre-determined state of readiness;

activating or deactivating said one or more computing devices, said activating or deactivating resulting in a number of computing devices in one or more states of readiness i;

computing an optimal configuration of said computing devices in said one or more said states of readiness i by minimizing an expected power consumption Σ

_i A_i P_i where i represents all inactive computing devices, A_i represents a number of said computing devices currently in each state i, and, P_i represents an average power consumption of computing devices while in said state i; and

placing said computing devices in said optimized configuration, and

formulating an objective function to be minimized, said objective function representing said expected power consumption Σ

_i A_iP_i subject to a constraint,

said optimizing including computing activation vector A″

representing a number of computing devices in said optimized one or more activation states of readiness i;

specifying an upper threshold T_max representing a maximum expected time to activate additional computing devices; and

computing probabilities p_i for i additional computing devices to be needed, where i ranges from 1 to a specified upper threshold, wherein said activation vector A″

is computed by formulating said objective function to minimize an expected power consumption Σ

_i A″

_i P_i given a constraint Σ

_iR_i p_i<

T_max, where Σ

_iR_ip_i is a weighted sum representing the expected time required to make the i computing devices available.

10. The computer program product as claimed in claim 9, wherein, after one of said activating or deactivating one or more computing devices:

computing a resulting activation vector A′

representing said resulting number of computing devices in one or more activation states of readiness i.

11. The computer program product as claimed in claim 10, wherein said placing said computing devices in said optimized configuration comprises:

transitioning said computing devices from one inactive state to another to reach a desired power state for each computing device, said transitioning comprising;

generating a plan to minimize a time required to accomplish said transitioning from said activation vector A′

to optimized activation vector A″

according to a transition latency matrix T_ij, representing an expected time required to change an activation state of a computing device from activation state i to activation state j.

12. The computer program product as claimed in claim 10, wherein said objective function is subject to a constraint that R_k≦

R⁰_k for each computing devices k, wherein said R represents a readiness vector, a k^th component R_k representing a minimal amount of time for k computing devices to be restored to an active state given a specific activation vector A and said transition latency matrix T; and

, R⁰ represents a specified readiness vector, a k^th component R⁰_k representing a specified constraint on a maximum time permitted for k inactive servers to be restored to said active state.

13. The computer program product as claimed in claim 10, further comprising:

specifying a desired demand fluctuation response vector F⁰ such that a k^th component F⁰_k represents a maximum time permitted for a plurality of computing devices to be transitioned in anticipation of a workload intensity increase above a current level; and

converting a requirement expressed in terms of said fluctuation response vector F0 to said readiness vector R⁰.

14. The computer program product as claimed in claim 10, further comprising:

computing said probabilities p_s for “

s”

computing devices by;

specifying c(w) that represents a cumulative workload intensity distribution as a function of a workload intensity “

w”

;

computing a maximal workload intensity level w(s) for which “

s”

servers can attain a pre-determined performance level w(s);

computing a corresponding cumulative distribution value c(w(s)) to denote a probability that an observed cluster-level workload will be less than or equal to w(s); and

computing probability p_s for “

s”

servers to be required according to;

p_s=c(w(s))−

c(w(s−

1))

where cumulative density values c(w(s)) and c(w(s−

1)) correspond respectively to attainable workload intensity for an amount s servers w(s) and the attainable workload intensity for an amount s−

1 servers.

15. The computer program product as claimed in claim 10, wherein, after calculating said desired activation vector A″

,

modifying the desired activation vector into a user-preferred Activation Vector A′

″

; and

inputting, to a planner device, a sequence of actions for transitioning the computing devices according to said user-preferred Activation Vector A′

″

.

16. A method for managing power consumption in a system including computing devices in both active and one or more inactive states of readiness, the method comprising:

receiving, at a processing unit, a request to one of;

activate or deactivate one or more computing devices from a respective group of inactive or active computing devices, each said inactive or active computing device initially in a pre-determined state of readiness;

activating or deactivating said one or more computing devices, said activating or deactivating resulting in a number of computing devices in one or more states of readiness i;

computing an optimal configuration of said computing devices in said one or more said states of readiness i by minimizing an expected power consumption Σ

_i A_i P_i where i represents all inactive computing device states, A_i represents a number of said computing devices currently in each state i, and, P_i represents an average power consumption of computing devices while in said state i; and

,

placing said computing devices in said optimized configuration,

formulating an objective function to be minimized, said objective function representing said expected power consumption Σ

_i A_iP_i subject to a constraint,

computing an activation vector A″

representing a number of computing devices in said optimized one or more activation states of readiness i;

specifying an upper threshold T_max representing a maximum expected time to activate additional computing devices; and

computing probabilities p_i for i additional computing devices to be needed, where i ranges from 1 to a specified upper threshold, wherein said activation vector A″

is computed by formulating said objective function to minimize an expected power consumption Σ

_i A″

_i P_i given a constraint Σ

_iR_i p_i<

T_max, where Σ

_iR_ip_i is a weighted sum representing the expected time required to make the i computing devices available,

wherein a program using said processor unit is configured to perform one or more of said receiving, activating, computing, placing, formulating and computing probabilities.

17. The method as claimed in claim 16, wherein, after one of said activating or deactivating one or more computing devices:

computing a resulting activation vector A′

representing said resulting number of computing devices in one or more activation states of readiness i.

18. The method as claimed in claim 17, further comprising:

wherein said placing said computing devices in said optimized configuration comprises;

transitioning said computing devices from one inactive state to another to reach a desired power state for each computing device, said transitioning comprising;

generating a plan to minimize a time required to accomplish said transitioning from said activation vector A′

to optimized activation vector A″

according to a transition latency matrix T_ij, representing an expected time required to change an activation state of a computing device from activation state i to activation state j.

19. The method as claimed in claim 17, wherein said objective function is subject to a constraint that R_k≦

R⁰_k for each computing devices k, wherein said R represents a readiness vector, a k^th component R_k representing a minimal amount of time for k computing devices to be restored to an active state given a specific activation vector A and said transition latency matrix T; and

, R⁰ represents a specified readiness vector, a k^th component R⁰_k representing a specified constraint on a maximum time permitted for k inactive servers to be restored to said active state.

20. The method as claimed in claim 17, further comprising:

specifying a desired demand fluctuation response vector F⁰ such that a k^th component F⁰_k represents a maximum time permitted for a plurality of computing devices to be transitioned in anticipation of a workload intensity increase above a current level; and

converting a requirement expressed in terms of said fluctuation response vector F⁰ to said readiness vector R⁰.

21. The method as claimed in claim 17, further comprising:

computing said probabilities p_s for “

s”

computing devices by;

specifying c(w) that represents a cumulative workload intensity distribution as a function of a workload intensity “

w”

;

computing a maximal workload intensity level w(s) for which “

s”

servers can attain a pre-determined performance level w(s);

computing a corresponding cumulative distribution value c(w(s)) to denote a probability that an observed cluster-level workload will be less than or equal to w(s); and

computing probability p_s for “

s”

servers to be required according to;

p_s=c(w(s))−

c(w(s−

1));

where cumulative density values c(w(s)) and c(w(s−

1)) correspond respectively to attainable workload intensity for an amount s servers w(s) and the attainable workload intensity for an amount s−

1 servers.

22. The method as claimed in claim 17, wherein said objective function is subject to a constraint that based on minimizing wear-and-tear on computing resources that result from changing their respective activation state, said constraint comprising one or more of:

specifying a total number of changes in activation state from exceeding a specified value, specifying a total number of changes in activation state within any given time period from exceeding a given upper limit.

23. The method as claimed in claim 17, wherein, after calculating said desired activation vector A″

,

modifying the desired activation vector into a user-preferred Activation Vector A′

″

; and

,

inputting, to a planner device, a sequence of actions for transitioning the computing devices according to said user-preferred Activation Vector A′

″

.

24. The method as claimed in claim 17, further comprising:

receiving, at said processor unit, a request to de-activate one or more active computing devices; and

placing said de-activated computing devices at a specified state of readiness, thereby, resulting in a new activation vector A′

;

solving a constraint optimization problem for computing a best activation vector A″

from said activation state vector A′

;

generating a plan that minimizes a sequence of actions required to switch state of said computing devices from activation state A′

to activation state A″

; and

executing the plan to bring said computing devices to said activation state A″

.