The fact that the memory is high is not necessarily a bad thing. Faster garbage collection implies more CPU usage.
I had some success following the tuning advice here, to make my memory usage less spikey http://blog.mikiobraun.de/2010/08/cassandra-gc-tuning.html Again, less spikes != better performance, is not a fact. On Tue, Sep 7, 2010 at 9:25 PM, shangan <shan...@corp.kaixin001.com> wrote: > how to change the configure in order to trigger GC earlier not when it is > close to the memory maximum? > > > 2010-09-08 > > > > shangan > > > > 发件人: Steve Loughran > 发送时间: 2010-09-06 18:16:51 > 收件人: common-user > 抄送: > 主题: Re: namenode consume quite a lot of memory with only serveral hundredsof > files in it > > On 06/09/10 08:27, shangan wrote: >> my cluster consists of 8 nodes with the namenode in an independent >> machine,the following info is what I get from namenode web ui: >> 291 files and directories, 1312 blocks = 1603 total. Heap Size is 2.92 GB / >> 4.34 GB (67%) >> I'm wondering why the namenode take so much memory while I only store >> hundreds of files. I've check the fsimage and edits files, the size of the >> sum of both is only 232 KB. So far as I know namenode can store the >> information of millions of files with 1G RAM, why my cluster consume so much >> memory ? If it goes on,I can't store that many files before the memory is >> eaten up. >> > It might just been there isn't enough memory consumption on your > pre-allocated heap to trigger GC yet; have a play with the GC tooling > and jvisualvm to see what's going on. > __________ Information from ESET NOD32 Antivirus, version of virus signature > database 5418 (20100902) __________ > The message was checked by ESET NOD32 Antivirus. > http://www.eset.com >