Posts by Ramuh


log in
1) Message boards : Number crunching : Crunching / Gaming Computer Build 2010 (Message 950790)
Posted 1603 days ago by Profile Ramuh
120mm 3000rpm fan? I hope you're not planning on sleeping in the same *house*. :)

Also, if those are the speakers you are going to use, you might as well use the integrated sound card.
2) Message boards : Technical News : Sad Trombone (Nov 25 2009) (Message 949895)
Posted 1606 days ago by Profile Ramuh
You're talking a lot about MySQL problems on that new monster of yours, and in case you weren't aware, I just thought you should know that MySQL scales horribly on a large number of cores. It can barely scale to 8, so there's virtually no chance that you're getting good results on 24.

Sun has an "official" writeup of a scaling attempt here: http://blogs.sun.com/mrbenchmark/entry/scaling_mysql_on_a_256

Their suggestion? Run several instances of MySQL on the same server. Which is a bit meh, but if you insist on running MySQL on a high-performance project like this, it might be worth looking into splitting up the most trafficked tables into separate instances. Then you could also enable the safe commit behavior individually on the critical servers.

You've also mentioned running into replica problems on jocelyn after a primary crash, which is very common. Here's a trick to get around it, at least temporarily.

After the crash, feed jocelyn the following through the MySQL console:

CHANGE MASTER TO MASTER_LOG_FILE=[NEXT FILE], MASTER_LOG_POS=4;
START SLAVE;

Replace [NEXT FILE] with the name of the first binary log file on mork that it started writing after the crash, typically mysql-bin.002342 or something similar. You can get the name by running SHOW MASTER STATUS; on mork. This will skip the corrupted end of the previous binary log, and restart replication from the new file.

Then verify that it's running with a SHOW SLAVE STATUS;

Note that you *cannot be sure* that the state on the replica and the primary is now consistent unless you are using safe commits, however it should still be consistent enough for non-science use.
3) Message boards : Number crunching : Welcome to the 10 year club (Message 949878)
Posted 1606 days ago by Profile Ramuh
So, I've been on-again off-again for months at a time (can't run this thing during summer), but yea. Hi. :)

As a sidenote, this i7 has been crunching more in two weeks than my previous contestant managed in 8 months.


Copyright © 2014 University of California