Generating predictions for host machine deployments
Disclosed are various embodiments for generating recommended replacement host machines for a datacenter. The recommendations can be generated based upon an analysis of historical workload usage across the datacenter. Clusters can be generated that cluster workloads together that are similar. Purchase plans can be generated based upon the identified clusters and benchmark data regarding servers.
1. A system comprising:
at least one computing device comprising at least one processor and at least one data store:
machine readable instructions stored in the at least one data store, wherein the instructions, when executed by the at least one processor, cause the at least one computing device to at least:
identify host data for a plurality of host machines in a data center, the host data identifying the host machines comprising the data center, the host data further identifying end-of-life information associated with respective ones of the plurality of host machines;
identify resource utilization data associated with the plurality of host machines, the resource utilization data comprising time series data identifying resource utilization by a plurality of workloads deployed across the host machines;
generate a plurality of clusters of workloads based upon the resource utilization data, the clusters generated by clustering workloads that are similar to each other based upon the utilization data;
generate respective usage predictions for the workloads based upon the resource utilization data;
generate a forecasted resource requirement for the clusters based upon the respective usage predictions, the forecasted resource requirement having a time horizon until a subsequent server upgrade;
generate a collective resource requirement for a plurality of replacement host machines based upon the forecasted resource requirement and the respective usage predictions;
identify benchmark data for a plurality of candidate replacement host machines to replace one or more of the host machines, the benchmark data comprising computing capabilities and a cost of respective candidate host machines;
generate a recommendation for the plurality of replacement host machines based upon the benchmark data and the collective resource requirement; and
cause the at least one computing device to at least map the workloads to respective one of the replacement host machines by identifying replacement host machine having a first ratio of resource parameters closest to a second ratio of the resource parameters defined by the respective usage prediction of the workloads.
2. The system of claim 1 , wherein the resource utilization data is identified by identifying at least one of a plurality of resource metrics, wherein the plurality of resource metrics are at least one of: a virtual central processing unit (vCPU) usage, a memory usage, a network input/output operations per second (IOPS), a network bandwidth usage, or a disk usage associated with the plurality of workloads deployed on the plurality of host machines.
3. The system of claim 2 , wherein the plurality of clusters of workloads are generated by identifying a respective median value of a plurality of resource metrics associated with respective ones of the workloads the clustering the workloads deployed on the host machines by the respective median values.
4. The system of claim 3 , wherein the clusters of workloads are generated by performing an unsupervised clustering algorithm on the respective median values of the plurality of resource metrics.
5. The system of claim 1 , wherein the respective usage predictions for the clusters are generated by performing a Holt's Forecasting model, wherein a first input into the model comprises the time-series data and a second input into the model comprises an expected time period of deployment of the replacement host machines.
6. The system of claim 5 , wherein the respective usage predictions further comprises a headroom parameter that increases the respective usage predictions beyond a usage forecasted by the model.
7. A method comprising:
identifying host data for a plurality of host machines in a data center, the host data identifying the host machines comprising the data center, the host data further identifying end-of-life information associated with respective ones of the plurality of host machines;
identifying resource utilization data associated with the plurality of host machines, the resource utilization data comprising time series data identifying resource utilization by a plurality of workloads deployed across the host machines;
generating a plurality of clusters of workloads based upon the resource utilization data, the clusters generated by clustering workloads that are similar to each other based upon the utilization data;
generating respective usage predictions for the workloads based upon the resource utilization data;
generating a forecasted resource requirement for the clusters based upon the respective usage predictions the forecasted resource requirement having a time horizon until a subsequent server upgrade;
generating a collective resource requirement for a plurality of replacement host machines based upon the forecasted resource requirement and the respective usage predictions;
identifying benchmark data for a plurality of candidate replacement host machines to replace one or more of the host machines, the benchmark data comprising computing capabilities and a cost of respective candidate host machines;
generating a recommendation for the plurality of replacement host machines based upon the benchmark data and the collective resource requirement; and
causing the at least one computing device to at least map the workloads to respective one of the replacement host machines by identifying replacement host machine having a first ratio of resource parameters closest to a second ratio of the resource parameters defined by the respective usage prediction of the workloads.
8. The method of claim 7 , wherein the resource utilization data is identified by identifying at least one of a plurality of resource metrics, wherein the plurality of resource metrics are at least one of: a virtual central processing unit (vCPU) usage, a memory usage, a network input/output operations per second (IOPS), a network bandwidth usage, or a disk usage associated with the plurality of workloads deployed on the plurality of host machines.
9. The method of claim 8 , wherein the plurality of clusters of workloads are generated by identifying a respective median value of a plurality of resource metrics associated with respective ones of the workloads the clustering the workloads deployed on the host machines by the respective median values.
10. The method of claim 9 , wherein the clusters of workloads are generated by performing an unsupervised clustering algorithm on the respective median values of the plurality of resource metrics.
11. The method of claim 7 , wherein the respective usage predictions for the clusters are generated by performing a Holt's Forecasting model, wherein a first input into the model comprises the time-series data and a second input into the model comprises an expected time period of deployment of the replacement host machines.
12. The method of claim 11 , wherein the respective usage predictions further comprises a headroom parameter that increases the respective usage predictions beyond a usage forecasted by the model.
13. A non-transitory computer-readable medium comprising machine readable instructions, wherein the instructions, when executed by at least one processor, cause at least one computing device to at least:
identify host data for a plurality of host machines in a data center, the host data identifying the host machines comprising the data center, the host data further identifying end-of-life information associated with respective ones of the plurality of host machines;
identify resource utilization data associated with the plurality of host machines, the resource utilization data comprising time series data identifying resource utilization by a plurality of workloads deployed across the host machines;
generate a plurality of clusters of workloads based upon the resource utilization data, the clusters generated by clustering workloads that are similar to each other based upon the utilization data;
generate respective usage predictions for the workloads based upon the resource utilization data;
generate a forecasted resource requirement for the clusters based upon the respective usage predictions, the forecasted resource requirement having a time horizon until a subsequent server upgrade;
generate a collective resource requirement for a plurality of replacement host machines based upon the forecasted resource requirement and the respective usage predictions;
identify benchmark data for a plurality of candidate replacement host machines to replace one or more of the host machines, the benchmark data comprising computing capabilities and a cost of respective candidate host machines;
generate a recommendation for the plurality of replacement host machines based upon the benchmark data and the collective resource requirement; and
cause the at least one computing device to at least map the workloads to respective one of the replacement host machines by identifying replacement host machine having a first ratio of resource parameters closest to a second ratio of the resource parameters defined by the respective usage prediction of the workloads.
14. The non-transitory computer-readable medium of claim 13 , wherein the resource utilization data is identified by identifying at least one of a plurality of resource metrics, wherein the plurality of resource metrics are at least one of: a virtual central processing unit (vCPU) usage, a memory usage, a network input/output operations per second (IOPS), a network bandwidth usage, or a disk usage associated with the plurality of workloads deployed on the plurality of host machines.
15. The non-transitory computer-readable medium of claim 14 , wherein the plurality of clusters of workloads are generated by identifying a respective median value of a plurality of resource metrics associated with respective ones of the workloads the clustering the workloads deployed on the host machines by the respective median values.
16. The non-transitory computer-readable medium of claim 15 , wherein the clusters of workloads are generated by performing an unsupervised clustering algorithm on the respective median values of the plurality of resource metrics.
17. The non-transitory computer-readable medium of claim 13 , wherein the respective usage predictions for the clusters are generated by performing a Holt's Forecasting model, wherein a first input into the model comprises the time-series data and a second input into the model comprises an expected time period of deployment of the replacement host machines.
18. The non-transitory computer-readable medium of claim 17 , wherein the respective usage predictions further comprises a headroom parameter that increases the respective usage predictions beyond a usage forecasted by the model.