It was the kernel...thanks for the help. We had considered but could not really believe a slight kernel version difference would cause this.
Thanks On Wed, Jan 18, 2012 at 11:47 PM, Stack <st...@duboce.net> wrote: > On Wed, Jan 18, 2012 at 8:40 AM, Wayne <wav...@gmail.com> wrote: > > We have set up a new cluster and are now seeing pdflush processes pegged > at > > 100% cpu during compaction. Is this normal? It sometimes "looks" like > > pdflush is pegged at 100% for quite a while and the java cpu usage goes > > down as if it is waiting on it to finish. We are using 0.90.4 on centos > 5.5 > > with no swap. Is this a linux kernal issue due to no swap? This cluster > is > > the first time we have ever seen this but we are running a slightly newer > > kernal than other cluster where we did not see this. > > > > I've not noticed it Wayne (can't say I was particularly on the lookout > for it but I suppose I'd have seen a 100%pdflush if it was going on). > > > Any ideas? Should we be worried about this? Is/can pdflush really be a > > bottleneck to disk writes? > > > > Could you try older kernel on one of the cluster members and see if it > still 100%s? > > What about the pdflush tunings on the two clusters? Are they the > same? (I was looking at this page: > http://www.westnet.com/~gsmith/content/linux-pdflush.htm) > > St.Ack >