Performance comparison inputs
Match observed conditions and scopes before drawing a bounded fit decision.
Compare startup observations under matched conditions
Keep cold and warm observations separate before averaging offers.
Compare payload shape as well as row count
Measure actual serialized bytes for two supplied equal-row workloads.
Define the interval measured by a response target
Keep server execution separate from the complete user wait.
Match test reads, writes and the shared record
Execute the supplied trace and check whether it represents the requested workload.
Check retained buffers after repeated task cycles
Measure explicit retained allocations after temporary work ends.
Release a worker slot after actual cancellation
Start queued work only after the cancelled local worker exits.