Re: A feature patch to NPH-ize existing CGI (CGI standard safe)

[Cliff Skolnick <cl...@organic.com>]

On Fri, 25 Aug 1995 16:12:28 EDT, rst@ai.mit.edu (Robert S. Thau) wrote:

} If Cardiff is SunOS (as I *think* I recall), then vmstat, iostat, and
} top can be useful for figuring out what it's up to (in particular,
} "vmstat -S 3" will give you a running report, on 3-second intervals,
} of ongoing swapping and paging activity).

But if it is Solaris, heh,  All I/O is considered paging which
makes figuring where all your I/O cycles are going a pain in the butt.
The only easy way I've found to do this on Solaris is to watch the load
increase until I see the CPU idle time go up.  At that time
some resource (disk I/O or memory) is in need of help.  Now there is
some sublty here and you need to make a good guess.  If you don't have
enough memory, you pound the disk so pick which one to increase
wisely...If you have lots of numbers you may see a spike in disk activity
when your load has increased linearly.  This is usually a sign of where
paging becomes a problem and it is memory.  If the disk load does not have
a spike it is probably disk bandwidth.  Buy more spindles or controllers.

As for apache, PI and PO should remain at about the same ratio.  If
you ever see the PO/PI ratio rise sharply buy more memory.  This is
assuming there is little writing going on (logs only) and that over
a few minutes the PI for I/O will equalize with log writing.  Any
additional increase in PI/PO is due to the system throwing stuff out
of memory.

While it is easy to guess wrong when your system is hosed totaly,
the odds are you needed both more disk bandwidth and memory anyway :)

Cliff