Mining pool latency and stale shares: how much distance actually costs
Educational content only, not financial advice. Figures below come from the published sources named in the text and were checked on 1 November 2026; hardware prices and network conditions change, so re-check before spending anything.
Ask a room of miners how to fix stale shares and someone will say: move closer to the pool server. For Bitcoin, that advice survives about thirty seconds of arithmetic. Latency is real, stale shares are real, and the connection between the two is much smaller than the folklore claims, while the reject reasons that actually cost a small miner money usually sit somewhere else entirely. This piece checks the maths from two independent sources, separates the four reject codes, and gives a diagnostic order that starts with the cheap, common causes.
What a reject rate actually measures
A reject rate is rejected shares divided by total shares submitted. The pool pays on accepted shares only, so every rejected share is electricity spent on work that earns nothing. A small rate, a fraction of a percent up to a percent or two, is normal on any pool. It becomes a problem when it climbs to several percent and stays there, or when the pool-side hashrate drifts noticeably below the hashrate the miner reports locally. Trust the pool-side numbers for revenue decisions; the miner's local tally never matches the pool's exactly.
Keep two counters separate. The reject rate is pool-side: shares sent and thrown out, almost always a networking, timing or configuration problem. The hardware error counter is different: bad nonces the control board catches before submission, pointing at a hot or failing hashboard, an aggressive overclock or a weak power domain. Chasing a network fix for a hardware counter wastes a weekend.
The latency maths, recomputed
A pool-operator analysis built on live network data puts same-region round trips at about 20ms and cross-continent round trips at about 200ms. Take the worst gap in that table, roughly 200ms of extra delay, and set it against the time between blocks. On a 300-second block time, 200 ÷ 300,000 = 0.067% of work wasted, the source's own figure, and it recomputes cleanly. On Bitcoin's roughly ten-minute average block interval the same delay is 200 ÷ 600,000 = 0.033%. For coins with block times of 60 seconds or more, the same analysis puts the cross-continent stale increase below 0.1%, and states the point plainly: latency does not reduce hashrate, because the miner keeps hashing while replies travel. The stale-share cost of distance is a rounding error next to normal pool luck variance.
| Extra round-trip delay | Block interval | Share of work exposed | Reading |
|---|---|---|---|
| 200ms | 300s | 0.067% | Negligible |
| 200ms | 600s (Bitcoin average) | 0.033% | Negligible |
| 300ms | 60s | 0.5% | Barely noticeable |
| 300ms | 15s | 2% | Latency genuinely matters |
The rule that falls out of the table: latency matters when it is a significant fraction of the block time. Bitcoin is at the far, forgiving end of that scale. That does not make the nearest regional stratum URL a bad choice, it costs nothing to pick it. It makes it a poor first suspect when a reject rate sits at 4%.
Four codes, four different problems
When a pool refuses a share over Stratum it returns a numeric error, and the code names the neighbourhood to search. Code 21, stale or job not found, is the most common: the share was valid, but the pool had already retired the job after a new block arrived. Code 23, low difficulty share, means the share missed the pool's current target, usually a difficulty-negotiation hiccup, a mid-session difficulty change the firmware handled poorly, or a flaky connection corrupting submissions. Code 22, duplicate share, points at a misbehaving proxy, a double-submitting firmware bug, or two workers colliding on the same credentials. Code 24, unauthorized worker, is a configuration error in the worker string. Only one of the four is primarily a latency story.
Where rejects are actually born
A share crosses three links between the miner and the pool's server. The local network is the most common and most fixable culprit: interference, a marginal Ethernet cable, a cheap switch, or a Wi-Fi bridge dropping packets. Miners belong on hardwired Ethernet, never Wi-Fi. The second link is the upstream route, unstable routing, packet loss and jitter on the path out of the building. The third is the pool server itself: overloaded, distant or misbehaving. The tell for the third is social, not technical: if many miners on the same pool spike at the same moment, the problem is upstream of any one rig.
A diagnostic order that starts cheap
1. Read the kernel log first. A log dominated by stale or job not found entries is a timing problem; low difficulty or duplicate entries point at configuration, firmware or a bad proxy.
2. Measure the path, not a speed test. From a machine on the same network, ping or trace the stratum host. Consistent low round-trip time is the goal; spikes and packet loss are the smoking gun.
3. Fix the physical link. Re-seat or replace the Ethernet cable, keep data cabling away from high-current DC leads, and bypass any powerline or Wi-Fi bridge in the path.
4. Use the nearest regional stratum URL the pool publishes. Free, sensible, and, per the maths above, a small lever on Bitcoin rather than a cure.
5. Check difficulty behaviour. Where the pool supports variable difficulty, let it settle; a fixed difficulty set far too low floods the link with shares and invites stales, while one set too high starves the statistics.
6. Change one variable at a time and give each change a few hours. Reject rates are noisy over short windows, and a five-minute sample will lie in either direction.
What success looks like: the pool-side reject rate settles back to a fraction of a percent and holds there over a full day, the pool-side hashrate converges with the locally reported hashrate, and the kernel log stops filling with stale entries.
A short decision rule
Reject rate under roughly 2% with pool-side hashrate tracking the miner? Leave it alone; the remaining loss is smaller than the variance you cannot control. Sustained rates in the 3 to 5% range or higher mean something on the three links is broken, work the procedure above before paying for proxies, new pools or new hardware. And if the reject reasons are minimal while the hardware error counter climbs, stop debugging the network entirely: that miner has a bench problem, not a pool problem.
, Dex Carter, Tech Writer