Benchmarking Results
Reproducible, cold-start performance studies on real workloads, published in full with numbers and methodology included.
Agent Sandbox Benchmark: Nirvana vs GKE vs E2B
Nirvana recovered all 5,501 committed checkpoints on a brand-new pod after the compute was killed mid-write, sustains 18,733 durable checkpoints/s at matched 8 vCPU, releases idle compute in 1.5 s, and held 2.71 GiB/s flat through a 30-minute run.
idle compute, against 9.6 s on GKE.
Read reportLangChain Agent Benchmark: checkpoint writes
Part 2 of the LangChain agent benchmark — the write path. Checkpoints that fsync after every agent step, 100K commits, across 5 storage platforms. ABS finishes first in all 18 runs, with half the Postgres commit p99 at sustained scale.
end-to-end at 100k checkpoints, every commit durable.
Read reportLangChain Agent Benchmark: cold reads
Part 1 of the LangChain agent benchmark — the read path. Cold vector search and checkpoint loads across 5 storage platforms, 100K tasks, 1,000 agents. ABS finishes first at every scale tested.
end-to-end task completion at every scale.
Read reportClickHouse® Star Schema Benchmark (SSB): Nirvana × Altinity
13 SSB queries on single-node ClickHouse 25.8.16 (Altinity-managed) at SF=100 and SF=500. Selective queries land at 34–63 ms against a 3-billion-row table; the full-scan family sustains ~70M rows/sec from disk.
selective-query warm latency against a 3-billion-row table.
Read reportClickBench: Nirvana ABS vs AWS EBS
43 OLAP queries under forced cold-read, raw FIO synthetic tests across a 16-VM cluster, and normalized price-performance against AWS gp3 and io2.
on cold-read analytical queries vs AWS gp3.
Read reportPowering AI, blockchain, and
databases
