Software Performance: Cache Locality And Allocation
jprahman.substack.com
Introduction In our first note we discussed the tail latency ↔ variance equivalence. We also outlined a simplistic model to decompose software into a sequence of operations, each such operation has an execution time distribution characterized by both an expected execution time + variance. The execution time for software then becomes a linear combination of expected execution times and variances of the sub-operations. Based on this knowledge, we understand there are two ways to reduce tail latency:
Software Performance: Cache Locality And Allocation
Software Performance: Cache Locality And…
Software Performance: Cache Locality And Allocation
Introduction In our first note we discussed the tail latency ↔ variance equivalence. We also outlined a simplistic model to decompose software into a sequence of operations, each such operation has an execution time distribution characterized by both an expected execution time + variance. The execution time for software then becomes a linear combination of expected execution times and variances of the sub-operations. Based on this knowledge, we understand there are two ways to reduce tail latency: