From: Scott Shealy (spshealy@YAHOO.COM)
Date: Wed Oct 22 2003 - 12:48:35 EDT
Background: IBM p690 - Aix 5.1
Database Enviroment with File based containers(not RAW) High IO Decision
Support enviroment charterized by lots of sequential IO
Recently we began monitoring fsbufwaincnt and xpagerbufwaitcnt. We have
mostly JFS2 filesystems(108) and we noticed that xpagerbufwaitcnt was
increasing rapidly. When we first looked at the counter the box had been up
18 days and had a value of 412677571. That gave us a rate of 2.3 millions
wait events per day. Which seems kind of high to us(no prior experience
with this cnt). So we incresed j2_nBufferPerPagerDevice from 512(default)
to 2048. After remounting the filesystems we still have the counter
increasing rather quickly in the millions per day. So we are contemplating
increasing it again but we are not sure how high to go. But we do have a
couple of questions:
1) How big an impact could the xpagerbufwaitcnt event have on the
throughput? Are we talking about a percentage point or two or are we
concievably talking about 50% or greater. We believe we have an IO
bottleneck somewhere... That is we are not getting the IO throughput that
we are expecting given the devices hooked up(8 Fast600s and a Shark F20)
2) Are there any negatives from increasing j2_nBufferPerPagerDevice other
memory usage(we have plenty... 64 GB). Does any one have any experience
with this vm parameter?
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 22:17:18 EDT