System restart today, February 17
Posted: Wed Feb 17, 2016 9:11 am
We've finally been able to complete the performance improvements that we've been working on for the past few weeks, and Chris certified them for release last night.
These improvements reduce the number of database queries to register 1000 shares from about 20,000 to 10. The queries also produce many fewer dead rows in the database, so VACUUM operations don't require long cleanups the next weekend.
At the same time, we've also implemented code to put these database operations on a separate core, and also to put daemon operations on a separate core as well. By placing these operations on a separate core, we were able to increase the number of miners who could run simultaneously after initial startup from about 250 to about 500. We haven't yet worked significantly on improving that number but plan to investigate that this weekend.
After the release, we expect that the database performance limitations will be eliminated, and the database will rarely, if ever, get behind again. The mining server is still CPU-limited, but that limit has not yet been reached.
The system will be offline for 5-10 minutes while Chris implements the changes.
These improvements reduce the number of database queries to register 1000 shares from about 20,000 to 10. The queries also produce many fewer dead rows in the database, so VACUUM operations don't require long cleanups the next weekend.
At the same time, we've also implemented code to put these database operations on a separate core, and also to put daemon operations on a separate core as well. By placing these operations on a separate core, we were able to increase the number of miners who could run simultaneously after initial startup from about 250 to about 500. We haven't yet worked significantly on improving that number but plan to investigate that this weekend.
After the release, we expect that the database performance limitations will be eliminated, and the database will rarely, if ever, get behind again. The mining server is still CPU-limited, but that limit has not yet been reached.
The system will be offline for 5-10 minutes while Chris implements the changes.