Machine learning-based power capping and virtual machine placement in cloud platforms

A system, including a processor, for deploying virtual machines to servers housed in respective chassis, the system configured to:

• apply a machine learning model to predict whether a request for deployment of a virtual machine corresponds to deployment of a user-facing virtual machine or a non-user-facing virtual machine;
• sort a list of candidate servers based on both a chassis score and a server score for each server in the list of the candidate servers to determine a ranked list of the candidate servers, wherein the server score depends at least on whether the request for the deployment of the virtual machine is a request for a deployment of a user-facing virtual machine or a request for a deployment of a non-user-facing virtual machine;
• and
• using the processor, deploy the virtual machine to a server with highest rank among the ranked list of the candidate servers.

A method, implemented by a processor, for deploying virtual machines to servers housed in respective chassis, the method comprising:

• predicting whether a request for deployment of a virtual machine corresponds to deployment of a user-facing virtual machine or a non-user facing virtual machine;
• using the processor, determining a chassis score for each server in a list of candidate servers for deployment of the virtual machine;
• using the processor, determining a server score for each server in the list of the candidate servers, wherein the server score is determined using a first method if the virtual machine is determined to be a user-facing virtual machine, and wherein the server score is determined using a second method, different from the first method, if the virtual machine is determined to be a non-user facing virtual machine;
• using the processor, sorting the list of the candidate servers based on both the chassis score and the server score to determine a ranked list of the candidate servers;
• using the processor, deploying the virtual machine to a server with highest rank among the ranked list of the candidate servers;
• assigning the user-facing virtual machine to a first group of cores associated with the server if the request for the deployment of the virtual machine is determined to be for a user-facing virtual machine, or else assigning the non-user-facing virtual machine to a second group of cores associated with the server;
• and
• lowering a frequency of the second group of cores in response to a first power alert associated with a chassis comprising the server.