<div dir="ltr"><div class="gmail_extra"><br /><div class="gmail_quote">On Wed, Jan 27, 2016 at 5:15 PM, Aleksander
Alekseev<span dir="ltr"><<a href="mailto:a.alekseev@postgrespro.ru"
target="_blank">a.alekseev@postgrespro.ru</a>></span>wrote:<br /><blockquote class="gmail_quote" style="margin:0px
0px0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div class="" id=":2o6"
style="overflow:hidden">Mostlikely HASHHDR.mutex is not only bottleneck in your case so my<br /> patch doesn't help
much.Unfortunately I don't have access to any<br /> POWER8 server so I can't investigate this issue. I suggest to use
a<br/> gettimeofday trick I described in a first message of this thread. Its<br /> time consuming but it gives a clear
understandingwhich code is keeping<br /> a lock.</div></blockquote></div><br />I have also tested the pgbench Readonly
testwhen data don't fit into shared buffer,<br />Because in this case HASHHDR.mutex access will be quite frequent.<br
/>Andin this case i do see very good improvement in POWER8 server.<br /><br />Test Result:<br />------------<br />Scale
Factor: 300 <br />Shared Buffer: 512MB <br />pgbench -c$ -j$ -S -M Prepared postgres<br /> <br
/>Client base patch<br />64 222173 318318<br />128 195805 262290<br clear="all" /><br />-- <br
/><divclass="gmail_signature"><div dir="ltr"><span style="color:rgb(80,0,80);font-size:12.8px">Regards,</span><br
style="color:rgb(80,0,80);font-size:12.8px"/><span style="color:rgb(80,0,80);font-size:12.8px">Dilip Kumar</span><br
style="color:rgb(80,0,80);font-size:12.8px"/><span style="color:rgb(80,0,80);font-size:12.8px">EnterpriseDB: </span><a
href="http://www.enterprisedb.com/"style="color:rgb(17,85,204);font-size:12.8px"
target="_blank">http://www.enterprisedb.com</a><br/></div></div></div></div>