Control device for estimation of power consumption and energy efficiency of application containers
Devices and techniques for estimating power consumption and efficiency of application containers configured to run on a server system, such as a cloud server, are provided. In an example, a method can include creating a benchmark application container, naming the benchmark application container on a host server, collecting power consumption information of the host server, collecting resource utilization information of the benchmark application container using an artificial workload, building a statistical model using the power consumption information and the resource utilization information, and generating a first power model of the host server.
1. A control device comprising:
a first interface to a host server;
a second interface to one or more application containers configured to run on the host server; and
processing circuitry configured to:
apply an artificial workload to a benchmark application container;
collect power consumption information of the host server;
collect resource utilization information of the benchmark application container;
build a statistical model using the power consumption information and the resource utilization information; and
generate a first power model of the host server,
wherein a first application container of the one or more application containers is a benchmark application container and a second application container is a real-time application container,
the resource utilization information includes information of the host server while running the benchmark application container including:
central processor utilization information (U CPU ),
disk input and output utilization information (U Disk ), and
memory utilization information (U Mem ), and
the processing circuitry is further configured to:
determine coefficients of the first power model for the host server using the central processor utilization information (U CPU ), the disk input and output utilization (U Disk ) and the memory utilization information (U Mem ) and power consumption of the host server while building the statistical model, the coefficients determined via regression modeling during a training phase using the benchmark application container, each coefficient indicating power consumption of a different component of the host server during the training phase, and
estimate the power consumption (P Ci ) of the real-time application container using:
P Ci =a CPU U CPU _ Ci +b Disk U Disk _ Ci +c Disk U Mem _ Ci +P H /n,
where i is an index associated with the real-time container application of a plurality of real-time container applications, n is the actual number of real-time container applications running on the host, P H is the idle power of the host, and a, b, c are the coefficients.
2. The control device of claim 1 , wherein
the processing circuitry is further configured to refine the coefficients during runtime by a comparison between a combined estimated power of each of the plurality of real-time container applications with a measured power of the host server and a reconciliation between the combined estimated power and the measured power based on the measured power being equal to the combined estimated power.
3. The control device of claim 1 , wherein the processing circuitry is further configured to collect an indicator of application-specific operations (useful_operations) performed by the real-time application container over an interval of time (t1−t2) using the second interface.
4. The control device of claim 3 , wherein the processing circuitry is further configured to provide an efficiency (E) of the real-time application container (f Ci ) using the estimate of the power consumption (P Ci ) and the indicator of application-specific operations, wherein
Ef
Ci
=
useful_operations
t
1
-
t
2
∫
t
2
t
1
P
c
(
t
)
.
5. The control device of claim 1 , wherein the processing circuitry is configured to determine the coefficients via linear regression.
6. A non-transitory machine-readable medium including instructions for estimating power consumption of a cloud computing application that, when executed on a machine cause the machine to perform operations configured to:
run a benchmark application container on a host server;
collect power consumption information of the host server;
collect resource utilization information of the benchmark application container using an artificial workload;
build a statistical model using the power consumption information and the resource utilization information; and
provide a first power model of the host server,
wherein the resource utilization information includes information of the host server while running the benchmark application container including:
central processor utilization information (U CPU ),
disk input and output utilization information (U Disk ), and
memory utilization information (U Mem ), and
the instructions cause the machine to perform operations configured to:
determine coefficients of the first power model for the host server using the central processor utilization information (U CPU ), the disk input and output utilization (U Disk ) and the memory utilization information (U Mem ) and power consumption of the host server while building the statistical model, the coefficients determined via regression modeling during a training phase using the benchmark application container, each coefficient indicating power consumption of a different component of the host server during the training phase, and
estimate the power consumption (P Ci ) of a real-time application container using:
P Ci =a CPU U CPU _ Ci +b Disk U Disk _ Ci +c Disk U Mem _ Ci +P H /n,
where i is an index associated with the real-time container application of a plurality of real-time container applications, n is the actual number of real-time container applications running on the host, P H is the idle power of the host, and a, b, c are the coefficients.
7. The machine-readable medium of claim 6 , further including instructions that, when executed on the machine, cause the machine to perform operations including operations to refine the coefficients during runtime by a comparison between a combined estimated power of each of the plurality of real-time container applications with a measured power of the host server and a reconciliation between the combined estimated power and the measured power based on the measured power being equal to the combined estimated power.
8. The machine-readable medium of claim 6 , further including instructions that, when executed on the machine, cause the machine to perform operations including collecting an indicator of application-specific operations (useful_operations) performed by a real-time application container over an interval of time (t1−t2).
9. The machine-readable medium of claim 8 , including instructions that, when executed on the machine, cause the machine to perform operations including providing an efficiency (E) of the real-time application container (f Ci ) using the estimate of the power consumption (P Ci ) and the indicator of application-specific operations, wherein
Ef
Ci
=
useful_operations
t
1
-
t
2
∫
t
2
t
1
P
c
(
t
)
.
10. The machine-readable medium of claim 6 , wherein the coefficients are determined via linear regression.
11. A method for estimating power consumption of a cloud computing application, the method comprising:
running a benchmark application container on a host server;
collecting power consumption information of the host server;
collecting resource utilization information of the benchmark application container using an artificial workload;
building a regression model using the power consumption information and the resource utilization information; and
generating a first power model of the host server using the regression model,
wherein the resource utilization information includes information of the host server while running the benchmark application container including:
central processor utilization information (U CPU ),
disk input and output utilization information (U Disk ), and
memory utilization information (U Mem ), and
the method further comprises:
determining coefficients of the first power model for the host server using the central processor utilization information (U CPU ), the disk input and output utilization (U Disk ) and the memory utilization information (U Mem ) and power consumption of the host server while building the statistical model, the coefficients determined via regression modeling during a training phase using the benchmark application container, each coefficient indicating power consumption of a different component of the host server during the training phase, and
estimating the power consumption (P Ci ) of a real-time application container using:
P Ci =a CPU U CPU _ Ci +b Disk U Disk _ Ci +c Disk U Mem _ Ci +P H /n,
where i is an index associated with the real-time container application of a plurality of real-time container applications, n is the actual number of real-time container applications running on the host, P H is the idle power of the host, and a, b, c are the coefficients.
12. The method of claim 11 , further comprising refining the coefficients during runtime by a comparison between a combined estimated power of each of the plurality of real-time container applications with a measured power of the host server and a reconciliation between the combined estimated power and the measured power based on the measured power being equal to the combined estimated power.
13. The method of claim 11 , including:
collecting an indicator of application-specific operations performed by a real-time application container over an interval of time; and
determining an efficiency of the real-time container application using the estimate of the power consumption (P Ci ) and the indicator of application-specific operations.
14. The method of claim 11 , wherein the coefficients are determined via linear regression.