Presign calculator · 2-of-3 · measured on GCP n2d-standard-16
Presigns: DKLS23 vs CGGMP21
Pick a protocol, a presign count, a vCPU count, a network speed and a topology. Times come from measured runs (a 100k DKLS run, 100 presign + 100 finish for both protocols at 4/8/16/32 vCPU, and cross-region runs); total time is linear in the count at c=1. Presignatures are single-use — every finish consumes one. Full data and method: silence-laboratories/threshold-bench.
RAM 64 GiB (4 GiB per vCPU)
—
offline time (presigns)
—
throughput
—
p50 per presign
—
wire traffic, active nodes
—
measured 100k DKLS: 1h 46m 38s, 0 errors
CPU 4 → 32 vCPU: p50 flat (62–65 ms DKLS · 2.64–2.65 s CGGMP)
RAM: 4 GiB per vCPU; measured runtime use ~0.6 GB
network floor at 1 Gbps: ~1 ms per presign — never the bottleneck
How to read it
- CPU: presign latency is flat from 4 to 32 vCPU (measured) because each party runs the protocol on a single core; more vCPUs buy concurrency, not lower per-op latency (concurrency is not modelled here).
- RAM: n2d-standard allocates 4 GiB per vCPU (16/32/64/128 GiB for 4/8/16/32). Measured runtime footprint was ~0.6 GB, so RAM is never the constraint for presigning.
- Network: the floor is bytes ÷ speed; at 1 Gbps it is ~1 ms per presign against ~64 ms of CPU, so datacenter network speeds are never the bottleneck. The old 20/10/5 Mbit tc sweep stays in the repo as sensitivity only.
- Cross-region latency was measured between europe-west3-a and europe-north1-a; byte figures are ens4 counters, including WebSocket/HTTP framing and message signing.
Get in touch
Unlock privacy preserving collaboration with Silence Laboratories
Book a Call© Silence Laboratories