Scale results

Generated 2026-09-18 from the scale campaigns. One host (14-core Apple silicon, 36 GB), all nodes on loopback: these are compute, broker and protocol ceilings, not wide-area figures. Full tables are in SCALE-RESULTS.md.

Headline

Area	Result
Federation size	240 sites on one host; solving placement and promoting takes 3 ms; the index uses 19 % of one core under heartbeats
Index	1,000,000 files ingested at 143,444 files/s; 1,811 MB heap; resolve p99 0.4 ms at 5,398 resolves/s; replica converged with an identical hash
Durability data path	Encode 233–291 MB/s, restore 254–393 MB/s for 256 MiB–1 GiB objects; every restore hash-verified
Loss detection and repair	Every lost holder found by audit in < 122 ms and regenerated without keys
Transport	Fragment push peaks at 577 MB/s node to node; client fetch 143.7 MB/s
Isolation	Control-plane RPC p99 0.72 ms while a 444 MB/s fragment flood runs
Storm	40 of 240 sites lost at once → 85 objects repaired in 27.2 s; 7 objects beyond the code's tolerance reported, not hidden; control p99 0.5 ms during the storm
Multi-region	72 sites across 8 bridged regions

Codec, with no fabric

Operation	Rate
Reed–Solomon (10,4), 14 threads	3,872 MB/s encode
Reed–Solomon (10,4), one thread	423–425 MB/s
AES-256-GCM decrypt	~3,400 MB/s per thread; 24,232 MB/s encrypt on 14 threads
SHA-256	2,318.9 MB/s
Shamir split / combine (n=9, t=5)	60 µs / 6 µs

Durability data path by object size

Representative rows (full grid in the design record):

Object	Code	Encode MB/s	Restore MB/s	Detect + repair
64 MiB	(10,4)	245.2	242.4	1.07 s
256 MiB	(3,2)	278.3	392.6	2.11 s
1 GiB	(10,4)	270.8	253.5	7.99 s

Index scale

Files	Ingest files/s	Heap MB	Resolve p50 / p99 ms	Replica equal
100,000	123,736	138	0.4 / 0.8	yes
250,000	142,501	304	0.2 / 0.4	yes
1,000,000	143,444	1,811	0.4 p99	yes

The replica must be sized like the core: a 1 GB replica ran out of memory at 10⁶ files.