T On Thu, Jun 17, 2010 at 4:53 PM, Peter Schuller <peter.schul...@infidyne.com> wrote: >> drive on an EBS volume). Does that pattern look bad or is it >> relatively normal? If it's not normal, any advice for what I should >> change to get better performance? > > Assuming you're running with the default cassandra VM options or > similar (mainly, the CMS GC), the graph looks fairly normal to me. You > would tend to expect frequent small variations that are the result of > young-generation GC:s (ParNew:s), with some less frequent larger dips > in memory usage associated with CMS completing a concurrent mark/sweep > phase, typically triggered by used heap size approaching the current > heap size. > > Normally the dips are somewhat indicative of the size of the actual > live set that your application is using; I'm not sure what's going on > in the middle of the graph with the dip being significantly higher. > Did you have very significant write activity during that period (but > it seems excessive even so). > > -- > / Peter Schuller
Thanks for looking at this guys. Peter, I added a columnfamily around that time which caused me to restart each node. One of my nodes had a hard time coming back to life because its small / partition was 100% full due to two large core dump files - which I'm looking into how to prevent (or at least change where they get dumped). I'm guessing the other two nodes experienced heavier loads during that time. -Mike -- Mike Subelsky oib.com // ignitebaltimore.com // subelsky.com @subelsky // (410) 929-4022
