Re: [RFC][PATCH] iowait statistics

[Bill Davidsen]

On Wed, 15 May 2002, William Lee Irwin III wrote:

> Boots compiles and runs on an 4-way physical HT box. I didn't wake
> the evil twins to cut down on the number of variables so it stayed
> 4-way despite the ability to go 8-way.
> 
> Sliding window of 120 seconds, sampled every 15 seconds, under a
> repetitive kernel compile load:
> 
> Wed May 15 09:56:37 PDT 2002
> cpu  60701 0 5137 203545 9327
> cpu0 15048 0 1566 50868 2298
> cpu1 15257 0 1176 50818 2392
> cpu2 15248 0 1346 50802 2247
> cpu3 15148 0 1049 51057 2390
	[... snip ...]
> Wed May 15 09:58:22 PDT 2002
> cpu  98583 0 8082 204779 9328
> cpu0 24538 0 2254 51205 2298
> cpu1 24521 0 2065 51180 2393
> cpu2 24704 0 1978 51230 2247
> cpu3 24820 0 1785 51164 2390
> 
> 
> It looks very constant, not sure if it should be otherwise.

You show-offs with your big memory and everything in it... Okay, boot that
puppy with mem=256m and try that again, particularly with -j4 (or -j8 with
HT on). I bet THAT will give you some IOwait!

I think you do want to try HT after you find out the memory is small
enough. Pure curiousity on my part, I assume it will work, although the
results might not be what I expect.

-- 
bill davidsen <[e-mail]>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.