+1 to what Mike said. I am running some Lucene benchmarks as we type and this is exactly what I just saw. On a beefy box with 32GB RAM I'm searching 63GB worth of Lucene indices. I gave the JVM 20GB (-Xmx20g) at first and saw a bit of disk IO. Then I lowered that max heap to 10GB and the disk IO disappeared! I increased the number of search threads from 16 to whatever 64x8 is, and still no IO. There is no Solr in this benchmark I'm doing, but the same ideas apply.
iostat and vmstat are your friends. Otis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Simpy -- http://www.simpy.com/ - Tag - Search - Share ----- Original Message ---- From: Mike Klaas <[EMAIL PROTECTED]> To: solr-user@lucene.apache.org Sent: Friday, March 16, 2007 4:23:35 PM Subject: Re: Commit after how many updates? On 3/16/07, Chris Hostetter <[EMAIL PROTECTED]> wrote: > > : I thought so, but hoped there would be some experiences with heap space > : settings for Solr. But I guess I have to try for myself. > > there's lots of experience, but it's hard to translate to generic rules > ... there's so many variables involved that it's hard to even recognize > what the equation is. > > My advice: throw as much ram as you've got at it, slam it with realistic > load, watch your GC logs/graphs and dial it back as much as you can > without hurting things.. I'd temper this by suggesting that you always leave a healthy amount for the OS disk cache as well--you definitely don't want Solr occupying _all_ the memory on a machine. -Mike
