Concurrency
How the network holds as parallel sessions ramp: c1 → c4 → c8 (c16 for the one provider that clears it).
metric: max concurrency cleanThe tests
Same script. Every provider. Every run.
Ten tests, seven live today. Each one produces a number that lands on every provider's report tab. Nothing hand-scored, no vibes.
Latest run snapshot
September 2026: 6 providers, 20 sites, ~2,400 requests each.
Site-mean headline numbers from the residential run. Each provider row links to its full report.
| # | Provider | Success | Timeout | p50 latency | Cost / 1k | Ceiling |
|---|---|---|---|---|---|---|
| 01 | decodo | 76.9% | 0.1% | 1,191 ms | $0.63 | c8 |
| 02 | dataimpulse | 76.8% | 2.8% | 2,114 ms | $0.15 | c16 |
| 03 | oxylabs | 72.4% | 0.3% | 1,180 ms | $0.86 | c8 |
| 04 | webshare | 71.7% | 0.5% | 1,242 ms | $0.51 | c8 |
| 05 | shifter | 70.1% | 4.5% | 2,402 ms | $0.33 | c8 |
| 06 | byteful | 69.3% | 1.4% | 961 ms | $0.46 | c8 |
Run: 2026-09-15 / 16 · residential proxies · sequential across providers · no retries · from OVERALL.json
What each test measures
Ten tests. Seven live. Three coming.
Live tests already produce numbers on every provider's report tab. Coming tests will earn their own detail page once the data lands.
IP rotation
Whether every request gets a fresh IP across a pool drain.
metric: unique IPs per 1kIP pool
Country and ASN distribution. Diversity and repetition.
metric: ASN + country spreadGeography
Does a proxy tagged as US actually egress from the US?
metric: geo match %Latency
Connect + response time under load. p50 and p95.
metric: ms per requestReliability
Success, block, and timeout across the full 20-site set.
metric: success / block / timeout %Vendor split
Success against the four bot-protection vendors (Akamai, Cloudflare, PerimeterX, unknown).
metric: success rate per vendorSticky sessions
Does a sticky session actually hold the same IP over the advertised duration?
metric: session hold rate · comingSustained throughput
Steady 5-minute run at c10 to catch slow-degradation the ramp misses.
metric: RPS over 5 min · comingHome-IP baseline lift
The same 20 sites, from my home IP, so every provider is compared against a real baseline.
metric: per-site lift vs baseline · comingHow to read this
The report tab has everything. The essay is the argument.
Every headline number on this site comes from one of the tests above. If you want the raw per-site rows, per-concurrency latency table, and geo / country splits, open any provider's Residential Report tab. If you want a human take on what those numbers mean, read the review tab of the same page.
Full methodology (target set, transport client, session shape, definitions) lives on the methodology page.