Profiling session to occur Sunday or Monday night
Posted: Sat Apr 23, 2016 8:46 am
On Sunday or Monday night, Chris will be conducting a "profiling" session of the production mining server.
Months ago, we spent a significant amount of time improving the performance of the database. The changes have proven successful, and we estimate that the database can now handle the entire litecoin network's hashrate. However, CPU usage of the mining server was not significantly addressed at the time, and it may max out as early as 75 GH/s.
We've profiled the mining server on the development environment. While we can simulate a lot of users there, we don't have the resources to be able to use a lot of coins at once and to find blocks. The only way to figure out how much impact those two things have on performance, and which lines can provide the most significant improvements, is to enable a profiling mode that writes every function call to a file.
Because we've never done this before on production, we don't know if CPU usage will reach 100% under profiling mode, which would cause rejected shares. I expect that five minutes of this will be long enough to gather the data we need, and then we can improve performance offline. We'll need to do a restart, then five minutes of profiling, and then another restart to restore things to normal.
We apologize for the inconvenience. The impact to profits should be minimal, but you may be alerted through notification E-Mails that your rejected share amounts rose or that your hashrate dropped for a few minutes.
Months ago, we spent a significant amount of time improving the performance of the database. The changes have proven successful, and we estimate that the database can now handle the entire litecoin network's hashrate. However, CPU usage of the mining server was not significantly addressed at the time, and it may max out as early as 75 GH/s.
We've profiled the mining server on the development environment. While we can simulate a lot of users there, we don't have the resources to be able to use a lot of coins at once and to find blocks. The only way to figure out how much impact those two things have on performance, and which lines can provide the most significant improvements, is to enable a profiling mode that writes every function call to a file.
Because we've never done this before on production, we don't know if CPU usage will reach 100% under profiling mode, which would cause rejected shares. I expect that five minutes of this will be long enough to gather the data we need, and then we can improve performance offline. We'll need to do a restart, then five minutes of profiling, and then another restart to restore things to normal.
We apologize for the inconvenience. The impact to profits should be minimal, but you may be alerted through notification E-Mails that your rejected share amounts rose or that your hashrate dropped for a few minutes.