Performance

What your S3 client sees on a deployment like the one you would get: every request over TLS, authenticated, metered, encrypted at rest and written to the record. Nothing is turned off to measure it. our statement

Setup. A dedicated deployment in GCP us-west1 (4 vCPU, 16 GB) · client: a VM in the same region, a stock AWS SDK and curl · files kept on the deployment (placement LOCAL) · release r95 · 2026-10-30 · three back-to-back runs; each figure is the median run’s.

Store and read

	4 KiB	64 KiB	1 MiB	16 MiB
Store (PUT, p50 / p99 ms)	10 / 29	11 / 27	21 / 42	134 / 157
Read (GET, first byte p50 · whole p99 ms)	6 · 10	6 · 14	7 · 21	17 · 89
Check it exists (HEAD, p50 / p99 ms)		6 / 8

A PUT returns once the file is stored on the deployment. Sealing to Walrus follows, on the file’s placement.

Large files

	Throughput (median)
Multipart upload, 256 MiB (16 MiB parts, 4 at a time)	101 MiB/s
Multipart upload, 1 GiB	102 MiB/s
Store a 16 MiB file, one PUT	119 MiB/s
Read a 16 MiB file, one stream	236 MB/s
Range read, 1 MiB from a 1 GiB file	first byte 17 ms · whole p99 30 ms

Under concurrency (64 KiB)

Clients	Read p50 / p99 ms	Store p50 / p99 ms
1	6 / 12	12 / 33
8	12 / 35	43 / 129

Each deployment admits requests up to a rate set for it; above that, a request is answered 429 and should be retried.

Run it yourself

Every figure comes from ordinary S3 calls. On a trial deployment, time the same calls with your own client: PUT and GET at the sizes above, and a multipart upload with 16 MiB parts.