Scaled-down load test models for testing real-world loads

A method, comprising:

• generating, by a processor, a test environment for a system under test, the system under test including a CPU, a memory, an input/output (I/O) subsystem, and a plurality of nodes that collectively execute a workload, wherein:
• the test environment comprises a virtual CPU configured to emulate a utilization behavior of the CPU, a virtual memory configured to emulate a capacity and access behavior of the memory, a virtual I/O subsystem configured to emulate a throughput behavior of the I/O subsystem, and a plurality of virtual nodes, each virtual node corresponding to a respective node in the plurality of nodes of the system under test,
• each virtual node functions under a virtual load similar to its corresponding node in the plurality of nodes functioning under a corresponding real-world load condition, and
• the test environment is generated based on data corresponding to a plurality of different load operations of the system;
• collecting performance metric data from the system under test across a plurality of different real-world load operations, the performance metric data including measured CPU utilization, memory utilization, and I/O throughput values associated with each of the plurality of nodes;
• executing a first machine learning algorithm trained using the collected performance metric data to control application of virtual loads to the virtual nodes by iteratively applying one or more different virtual loads to each virtual node in the test environment and automatedly adjusting the virtual loads by constraining values of the virtual loads until virtual resource utilization of a particular scaled-down load test model converges with the utilization behavior of the CPU, the capacity and access behavior of the memory, and the throughput behavior of the I/O subsystem under a pre-defined real-world load condition, wherein:
• each constrained value of the virtual loads applied to each virtual node is comparatively smaller relative to each of one or more corresponding real-world loads for each node of the plurality of nodes defining the pre-defined real-world load condition for the system, and
• each virtual load of the one or more different virtual loads applied to each virtual node is based on a specific constraint corresponding to each type of virtual node;
• and
• executing the particular scaled-down load test model to identify resource-specific undesirable conditions experienced by the system under test to enable modification of the system prior to exposing the system to real-world load conditions.

A computer program product comprising a non-transitory computer-readable storage medium configured to store code executable by a processor, the executable code, when executed by the processor, causes the processor to:

• generate a test environment for a system under test, the system under test including a CPU, a memory, an input/output (I/O) subsystem, and a plurality of nodes that collectively execute a workload, wherein:
• the test environment comprises a virtual CPU configured to emulate a utilization behavior of the CPU, a virtual memory configured to emulate a capacity and access behavior of the memory, a virtual I/O subsystem configured to emulate a throughput behavior of the I/O subsystem, and a plurality of virtual nodes of the system under test, each virtual node corresponding to a respective node in the plurality of nodes,
• each virtual node functions under a virtual load similar to its corresponding node in the plurality of nodes functioning under a corresponding real-world load condition, and
• the test environment is generated based on data corresponding to a plurality of different load operations of the system;
• collect performance metric data from the system under test across a plurality of different real-world load operations, the performance metric data including measured CPU utilization, memory utilization, and I/O throughput values associated with each of the plurality of nodes,
• execute a first machine learning algorithm trained using the collected performance metric data to control application of virtual loads to the virtual nodes by iteratively applying one or more different virtual loads to each virtual node in the test environment and automatedly adjusting the virtual loads by constraining values of the virtual loads until virtual resource utilization of a particular scaled-down load test model converges with the utilization behavior of the CPU, the capacity and access behavior of the memory, and the throughput behavior of the I/O subsystem under a pre-defined real-world load condition, wherein:
• each constrained value of the virtual loads applied to each virtual node is comparatively smaller relative to each of one or more corresponding real-world loads for each node of the plurality of nodes defining the pre-defined real-world load condition for the system, and
• each virtual load of the one or more different virtual loads applied to each virtual node is based on a specific constraint corresponding to each type of virtual node;
• and
• execute the particular scaled-down load test model to identify resource-specific undesirable conditions experienced by the system under test to enable modification of the system prior to exposing the system to real-world load conditions.