Part of #151.
Problem. loadgen is amd64-only; both arches are load-driven from the amd64 client (msa2-client). Driving the arm64 server (msr1) from an amd64 client over the fabric is fine for throughput but mixes arch characteristics and serializes the two arches (×2 wall, hurts the 24h budget).
Goal. Symmetric load generation per arch for fidelity + parallelism.
Scope.
- Build/stage an arm64 loadgen binary.
- Use loadgen's federation (or run two independent loadgen instances + merge HdrHistograms via
report.mergeHistograms) so arm64 server is driven by arm64 and amd64 by amd64.
- loadgen federation is currently 2-node single-peer; assess whether it suffices or whether independent-runs-then-merge is simpler.
- Enables running both arches concurrently → roughly halves wall-clock.
Acceptance. arm64 results produced with an arm64 load driver; both arches run concurrently within budget.
Refs: mage_cluster.go:129-132 (v1.5 backlog note), loadgen federation.go, report/aggregate.go (mergeHistograms).
Part of #151.
Problem. loadgen is amd64-only; both arches are load-driven from the amd64 client (msa2-client). Driving the arm64 server (msr1) from an amd64 client over the fabric is fine for throughput but mixes arch characteristics and serializes the two arches (×2 wall, hurts the 24h budget).
Goal. Symmetric load generation per arch for fidelity + parallelism.
Scope.
report.mergeHistograms) so arm64 server is driven by arm64 and amd64 by amd64.Acceptance. arm64 results produced with an arm64 load driver; both arches run concurrently within budget.
Refs:
mage_cluster.go:129-132(v1.5 backlog note), loadgenfederation.go,report/aggregate.go(mergeHistograms).