Identifying hotspots and coldspots in forecasted power consumption data in an it data center for workload scheduling
Systems and methods are provided for using historic input power periodic data from a server in an IT data center to train a machine learning (ML) model to obtain forecasted power consumption data of the server for a future time period. Time windows of hotspots or coldspots are then identified in the forecasted power consumption data, hotspots being defined as areas or regions of over-utilization in a time series data, and coldspots being defined as areas or regions of under-utilization in a time series data. The hotspots and coldspots are identified by calculating an exponential mean average (EMA) of the forecasted power consumption data, taking points above the EMA as hotspots and points below the EMA as coldspots. The identified hotspots and coldspots can be used to schedule workloads for a server or a data center, to more efficiently plan existing workloads, or to introduce new workloads at more optimal time periods.
1 . A computing component, comprising:
at least one processor; and
a memory unit operatively connected to the at least one processor, the memory unit including instructions that, when executed, cause the at least one processor to:
train a machine learning model to obtain forecasted power consumption data of a server for a next future time period with at least one dataset, received from the server, of input periodic data from a past time period;
calculate an exponential mean average (EMA) of the forecasted power consumption data;
compare points of the forecasted power consumption data to the EMA;
identify consecutive points of forecast power consumption above the EMA as a time window of hotspots and consecutive points of forecast power consumption below the EMA as a time window of coldspots; and
schedule a workload to be executed during the time window of coldspots.
2 . The computing component of claim 1 , wherein the memory unit includes instructions that when executed further cause the at least one processor to identify a time window as a time window of hotspots only when all points of forecast power consumption in the time window are above the EMA, and identify a time window as a time window of coldspots only when all points of forecast power consumption in the time window are below the EMA.
3 . The computing component of claim 1 , wherein the memory unit includes instructions that when executed further cause the at least one processor to schedule a reduced workload to be executed during the time window of hotspots.
4 . The computing component of claim 1 , wherein the input periodic data is received from an Active Health System Log of the server.
5 . The computing component of claim 1 , wherein the server comprises the computing component.
6 . A computer-implemented method, comprising:
training a machine learning model to obtain forecasted power consumption data of a server for a next future time period with at least one dataset of input periodic data from a past time period;
calculating an exponential mean average (EMA) of the forecasted power consumption data;
comparing points of the forecasted power consumption data to the EMA;
identifying consecutive points of forecast power consumption above the EMA as a time window of hotspots and consecutive points of forecast power consumption below the EMA as a time window of coldspots; and
scheduling a workload to be executed during the time window of coldspots.
7 . The computer-implemented method of claim 6 , wherein the points of the forecasted power consumption are approximately hourly.
8 . The computer-implemented method of claim 7 , wherein a number of consecutive points of forecast power consumption in the time window of hotspots is 5.
9 . The computer-implemented method of claim 7 , wherein a number of consecutive points of forecast power consumption in the time window of coldspots is 5.
10 . The computer-implemented method of claim 6 , further comprising reducing a planned workload to be executed during the time window of hotspots.
11 . The computer-implemented method of claim 6 , further comprising receiving the at least one dataset of input periodic data of the server.
12 . The computer-implemented method of claim 6 , wherein the input periodic data is received from an Active Health System Log of the server.
13 . A non-transitory machine-readable storage medium comprising instructions executable by a processor, the instructions programming the processor to:
receive at least one dataset of input periodic data of a server;
using a trained machine learning model to obtain forecasted power consumption data of the server for a next future time period, the machine learning model being trained with the at least one dataset of input periodic data from a past time period;
calculating an exponential mean average (EMA) of the forecasted power consumption data;
comparing points of the forecasted power consumption data to the EMA;
identifying consecutive points of forecast power consumption above the EMA as a time window of hotspots and consecutive points of forecast power consumption below the EMA as a time window of coldspots; and
scheduling a workload to be executed during the time window of coldspots.
14 . The non-transitory machine-readable storage medium of claim 13 , wherein the points of the forecast power consumption are approximately hourly.
15 . The non-transitory machine-readable storage medium of claim 14 , wherein a number of consecutive points of forecast power consumption in the time window of hotspots is 5.
16 . The non-transitory machine-readable storage medium of claim 14 , wherein a number of consecutive points of forecast power consumption in the time window of coldspots is 5.
17 . The non-transitory machine-readable storage medium of claim 13 , the instructions further programming the processor to schedule a reduced workload to be executed during the time window of hotspots.
18 . The non-transitory machine-readable storage medium of claim 13 , wherein the input power periodic data is received from an Active Health System Log of the server.
19 . The computer-implemented method of claim 13 , wherein before the received at least one dataset of input power periodic data of the server is inputted into the machine learning model the at least one dataset is processed to interpolate any missing values from the at least one dataset and to detect and remove outliers in the at least one dataset.
20 . The computer-implemented method of claim 13 , wherein the machine learning model is one of SARMA, FBProphet, Holt-Winter-ES, or Gated Recurrent Unit Network.