I've just found that putting the server's .jar file directly into /dev/shm and running it from there results in a very significant increase in performance on my micro instance. It makes sense given how slow ebs stores for micro instances are.
I didn't need to use cpulimit, but then I was only testing it with myself logged in. Even then I was only seeing about 2% cpu usage. Does the server load go up more than you'd expect with more users? If you do want to make sure you get consistent performance, then cpulimit would make sure that server performance degrades predictably with more users.
I should point out that if you're doing this, then you need to make your own arrangements for backing up your world files etc. Especially in my case, as I'm running a spot request instance. (I only pay $6 a month on average, but my server is sometimes shutdown when prices go over US$0.01 / hr, and restarted when they come back down).
0
I didn't need to use cpulimit, but then I was only testing it with myself logged in. Even then I was only seeing about 2% cpu usage. Does the server load go up more than you'd expect with more users? If you do want to make sure you get consistent performance, then cpulimit would make sure that server performance degrades predictably with more users.
I should point out that if you're doing this, then you need to make your own arrangements for backing up your world files etc. Especially in my case, as I'm running a spot request instance. (I only pay $6 a month on average, but my server is sometimes shutdown when prices go over US$0.01 / hr, and restarted when they come back down).