Congestion-based network optimization for gpu communication

A non-transitory computer-readable media storing computer instructions that, when executed by a management processor in a cloud management platform, cause the management processor to:

• identify one or more workloads in a data center network based on analysis of network flow data and virtualization information collected from a plurality of network switches;
• monitor network latency across the plurality of network switches for data flows associated with the identified workloads;
• detect network congestion using statistical analysis of the monitored network latency;
• dynamically determine network optimization parameters in response to the detected congestion;
• and
• implement the network optimization parameters across one or more of the plurality of network switches to mitigate congestion for the workload.

A computer-implemented method for optimizing network performance for a workload, the computer-implemented method comprising:

• identifying, by a management processor in a cloud management platform, one or more workloads in a data center network based on analysis of network flow data and virtualization information collected from a plurality of network switches;
• monitoring, by the management processor, network latency across the plurality of network switches for data flows associated with the identified workloads;
• detecting, by the management processor, network congestion using statistical analysis of the monitored network latency;
• dynamically determining, by the management processor, network optimization parameters in response to the detected congestion;
• and
• implementing, by the management processor, the network optimization parameters across one or more of the plurality of network switches to mitigate congestion for the workload.