BenchmarksCopy Anchor Link

Live benchmarks use public mainnet endpoints, so results vary by machine, network, time, and provider rate limits. They are comparisons from one environment, not provider SLAs or load-test guarantees. Run them conservatively.

Sustained comparisonCopy Anchor Link

The optimized transport was compared with a direct PublicNode connection using ten concurrent workers for 15 seconds. Each worker requested getSlot, waited 100 ms after each call, and kept at most one request in flight. The pool used one retry with cache, coalescing, hedging, and health probes disabled so the measurement covers network routing rather than cache hits.

The earlier router reached a 383.77 ms p50 and 2,994.63 ms p95 under the same workload. Removing speculative in-flight penalties, keeping unknown providers off the healthy user path, and disabling PublicNode HTTP batching removed that overhead.

Rate-limit failoverCopy Anchor Link

This read-only edge case uses ten concurrent getSlot workers for 15 seconds against the official Solana RPC and then Smart Solana. No errors are injected and no transactions are sent. The pool uses one retry with cache, coalescing, hedging, and health probes disabled.

The official endpoint returned its first 40 successes during the initial second, then returned only HTTP 429 responses for the remaining 521 attempts. Its latency columns primarily measure error responses. Smart Solana routed the same sustained workload through healthy providers without a failed user request.

Run it from the repository root:

bun run bench:failover

The full per-second report is written to packages/smart-solana/failover-results.json.

Reproduce the PublicNode latency comparisonCopy Anchor Link

From the repository root:

BENCH_REGISTRY=1 \
BENCH_PROVIDER=publicnode \
BENCH_CONCURRENCY_LEVELS=10 \
BENCH_OUT=sustained-optimized-results.json \
bun run bench:sustained

BENCH_STAGE_MS, BENCH_WORKER_DELAY_MS, and BENCH_CONCURRENCY_LEVELS control the load. Omit BENCH_PROVIDER to compare every default provider. Set BENCH_REGISTRY=0 for the official Solana endpoint comparison.

Read-mix comparisonCopy Anchor Link

BENCH_OUT=benchmark-results.json bun run bench:live

This runner compares the direct public Solana RPC, an uncached pool, and the default cached pool over rotating sequential and concurrent phases. It reports cold and warm percentiles, errors, cache hits, hedges, attempts, and per-method results. BENCH_SEQUENTIAL, BENCH_CONCURRENT, and BENCH_PACE_MS adjust the workload.