Be careful of your NUMA boundaries. 20-30% is exactly the cross-bank access cost.
That's what I thought too. There are a lot of weird problems that can happen due to NUMA architecture issues. Reading the numactl for me really drives home just how much I don't understand it.
That's what I was thinking. Need some analysis on it really. We throw money at the problem at the moment and it goes away :)
Re:Disaster: MySQL 5.5 Flushing
by zwischenzug on Sunday September 18, 2011 @05:54PM
(#3011375)
I've seen DBs run on 128G machines on x6 and have no experience of such problems. Do you have a link or hard evidence of this?
I have seen throughput issues on VMWare machines (and false reporting on standard linux utilities, eg "time" output and various other issues that make me distrust it for performance-sensitive software), but don't know if this is to do with 128G machines; usually it's to do with contention on one network card handling all traffic to disks.
Re:Disaster: MySQL 5.5 Flushing
by trafficlight on Sunday September 18, 2011 @05:31PM
(#3011320)
I haven't heard/experienced the 36 GB thing since I don't run anything that big right now. Are you aware of any websites or blog posts that talk about it?
Some bits buried in here: http://frankdenneman.nl/2010/02/sizing-vms-and-numa-nodes/
I'd personally rather buy high end SPARC64 kit which has linear performance scalability but we all know what happened to Sun, plus it doesn't run Windows anyway :( whimpers a little