- Steve Sokolowski
- Archived posts: 4559
- Joined: Aug 27, 2014
Chris to issue share corrections
Chris will be issuing share corrections today for a period of time when miners were not credited balances. He should make the corrections this afternoon.
This morning's issues allowed us to make a breakthrough. We believe that many of the issues might be related to the way that coin assignment is handled when miners connect to the pool. We plan to make a change where there is a "connect loop," so that every second all miners who connect are assigned coins at once. The key to figuring this out was realizing that earlier we had made the number of "workers" at the top of the page a cached value, so it isn't updated as often as the hashrates. Therefore, it isn't obvious from the live data when miners connect or disconnect.
We had been trying to figure out why there were bursts of WAMP data, and we think that the reason is because sometimes there are surges of many miners connecting within a second, probably due to network conditions. By simply waiting to reply to new miners for a second and doing so all at once, we can send one WAMP update and reduce bandwidth usage if this is actually the cause.
This problem might also be responsible for some odd coin assignment errors that we were never able to figure out, and also why CPU usage tends to have spikes independent of server load.
This morning's issues allowed us to make a breakthrough. We believe that many of the issues might be related to the way that coin assignment is handled when miners connect to the pool. We plan to make a change where there is a "connect loop," so that every second all miners who connect are assigned coins at once. The key to figuring this out was realizing that earlier we had made the number of "workers" at the top of the page a cached value, so it isn't updated as often as the hashrates. Therefore, it isn't obvious from the live data when miners connect or disconnect.
We had been trying to figure out why there were bursts of WAMP data, and we think that the reason is because sometimes there are surges of many miners connecting within a second, probably due to network conditions. By simply waiting to reply to new miners for a second and doing so all at once, we can send one WAMP update and reduce bandwidth usage if this is actually the cause.
This problem might also be responsible for some odd coin assignment errors that we were never able to figure out, and also why CPU usage tends to have spikes independent of server load.
