Addressing CPU issues

I was able to significantly reduce CPU usage on several of the mining servers. It turns out that there are a few misconfigured miners that try to login thousands of times per minute, and dropping their packets upstream resolves the CPU usage issue for all but one of the servers.

With the other server, there are a lot of connections from a cloud mining service. Chris is going to work on a better load balancing system so that the connections are better distributed across the servers to improve performance for users of that cloud mining service at 209.216.176.34.

With the reduced CPU usage, stale shares are likely to improve for miners who are not using that service. Feel free to share your experiences.

for the last 2 hours i been kicked to my backup pool.

mycide wrote:for the last 2 hours i been kicked to my backup pool.

Are you using a shared connection with anyone?

No, it's my private internet connection at home.
Trying with a reboot on one of the L3+ it is not even connecting to PH, ui say _never_ connected to PH i have not had that shown before, have i been banned from connecting?

What is your IP address? I'll take a look.

Or, you can submit a support ticket with the address.

Chris made some changes to address this issue without needing to ban anyone.

Some of the mining servers had more connections than others, because we simply took the modulo of the last subnet of the IP addresses and assigned them to a mining server. We needed a repeatable algorithm for load balancing back in the time when the system was unstable and we couldn't depend upon the mining servers all starting at the same time.

Now, the algorithm is to assign new connections to a mining server at random. That means that load balancing will be equal, assuming that one mining server doesn't go offline and cause people to reconnect to the other servers, before coming back online itself with fewer connections. With the increased stability of the mining servers recently, everything seems to be working better than yesterday, so far. Currently, there are no blocked connections (because that code is disabled) and CPU load is around 50%, down from maximum on two of the servers yesterday.