TLUG Mailing List

Re: [tlug] Re: /proc/uptime

On Fri, Nov 14, 2008 at 11:51 AM, Stephen J. Turnbull <stephen@example.com> wrote:

Nguyen Vu Hung writes:

> > No. It indicates the amount of time any one of the cores has
> > nothing to do.

> Is this a good design? I think no. I will replace OR with AND.
> Not every recently applications are SMP aware, and some never is.

So? That doesn't change the fact that some of the processors are
doing almost no useful work!

I did mean "It indicates the amount of time any one of the cores has nothing to do"
does not make a good sense and it should be
"It indicates the amount of time *all the* cores has nothing to do".

And multi-thread programming (like
explicit heap management) is a waste of programmer skull-sweat in most
server applications as far as I can see; there's very little loss to
running multiple processes. For example, it's trivial to configure
one web proxy (which will run on one core) to front-end to several
webservers. That will get you core usage up (although if your disk IO
channels are already fully-loaded, it won't change the idle time %-age).

I take graphicsmagick as an example.
http://www.graphicsmagick.org/OpenMP.html
Before 1.3, it doens't support SMP so the gm command only runs on *ONE* core.
In 1.3, I can see ~x2 speedup on my dual Xeon linux box.

I don't need SMP support for "ls" command because it is fast enough.

So, if your number says some of your processors are idle 99% of the
time, and that bothers you, you're buying too many processors (in some
sense, of course there's an integer problem here). Either spend more
money on memory and IO channels (and get more output per box with the
same CPUs), or spend less on processors (and spend less per box).

I dual boot the dual Xeon box. On Windows, I run Adobe PS3, which is using Adobe Camera Raw)
to convert hundreds of Nikon NEF files to JPEG. I really benefit from SMP in this case.

vuhung16plus{remove}@example.com

http://www.hn.is.uec.ac.jp/~vuhung/Nguyen.Vu.Hung.html