Correlations between workload characteristics and elapsed times

A non-transitory machine-readable storage medium comprising instructions that upon execution cause a system to:

• receive a first collection of tokens relating to characteristics of workloads for a computing system;
• encode, by an encoder, the first collection of tokens to compute first weights representing relationships among tokens of the first collection of tokens;
• generate, by the encoder, a representation of the first collection of tokens based on the first weights;
• decode, by a decoder, a second collection of tokens to compute second weights representing relationships among tokens of the second collection of tokens, the second collection of tokens relating to elapsed times in executing the workloads;
• generate, by the decoder, a representation of the second collection of tokens based on the second weights;
• apply an attention function based on the representation of the first collection of tokens and the representation of the second collection of tokens, to produce third weights representing relationships among tokens of the first collection of tokens and tokens of the second collection of tokens;
• and
• determine, based on the third weights, relationships between the elapsed times and the characteristics of the workloads for resource allocation of resources of the computing system when executing the workloads.

A system comprising:

• a processor;
• and
• a non-transitory storage medium storing instructions executable on the processor to:
• receive, at an encoder, a first collection of tokens relating to characteristics of requests that produce workloads for a computing system;
• compute, using the encoder, first weights representing relationships among tokens of the first collection of tokens;
• generate, with the encoder, a representation of the first collection of tokens based on the first weights;
• receive, at a decoder from the encoder, the representation of the first collection of tokens;
• compute, using the decoder, second weights representing relationships among tokens of a second collection of tokens relating to elapsed times in executing the workloads;
• generate, with the decoder, a representation of the second collection of tokens based on the second weights;
• apply an attention function based on the representation of the first collection of tokens and the representation of the second collection of tokens, to produce third weights representing relationships among tokens of the first collection of tokens and tokens of the second collection of tokens;
• and
• determine, at the decoder based on the third weights, relationships between the elapsed times and the characteristics of the requests for resource allocation of resources of the computing system when executing the workloads.

A method comprising:

• receiving, at a system comprising a hardware processor, a first collection of tokens relating to characteristics of workloads for a computing system;
• encoding, by an encoder in the system, the first collection of tokens to compute first weights representing relationships among tokens of the first collection of tokens;
• generating, by the encoder, a representation of the first collection of tokens based on the first weights;
• decoding, by a decoder in the system, a second collection of tokens to compute second weights representing relationships among tokens of the second collection of tokens, the second collection of tokens relating to elapsed times in executing the workloads;
• generating, by the decoder, a representation of the second collection of tokens based on the second weights;
• applying, by the system, an attention function based on the representation of the first collection of tokens and the representation of the second collection of tokens, to produce third weights representing relationships among tokens of the first collection of tokens and tokens of the second collection of tokens;
• and
• determining, by the system based on the third weights, relationships between the elapsed times and the characteristics of the workloads for resource allocation of resources of the computing system when executing the workloads.