Re: page faults with phys. io?

From: Jason Riedy <ejr@dont-contact.us>
Date: Mon, 26 Jan 1998 00:19:49 -0500

Oh well. And Dirk Vleugels writes:
 -
 - This machine is running solaris 2.5.1, and as far as i remember solaris
 - will use the page mechanism to do all disk writes, correct?

I just started running 1.2.beta11 on a new Solaris 2.6 machine with more
than enough memory (about 7 times as much physical memory as I've devoted
to squid). I seem to be seeing something similar, but I'm generating
artificial loads.

Hmmm... I'm snarfing (slowly) Yahoo as a test. The cachemgr.cgi interface
is reporting 151 page faults with physical i/o out of 717 requests. Each
page-outs I've seen with vmstat is accompanied by a page-free. That's
consistent with flushing unread, cached writes. I'm getting three page-outs
every 15-20 seconds or so. That seems to match wget's rate, but I don't
feel like making calculations. ;) Nothing else is really running on
the system right now.

If you even think it's possible you don't have enough memory, watch the
vmstat output. The Performance Q&A column in SunWorld has a wonderful
explanation of its output at
        http://www.sun.com/sunworldonline/swol-05-1997/swol-05-perf.html .
That column also has a bit on the filesystem cache.

Jason, wannabe optimizer...
Received on Sun Jan 25 1998 - 21:31:25 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:38:30 MST