Subject: Bogomips, WAS Re: wlan-ng fix?
From: Christopher E. Brown (cbrown@woods.net)
Date: Mon Mar 03 2003 - 05:45:06 AKST
On 3 Mar 2003, Jim Gribbin wrote:
> BTW - I was comparing notes with a couple of people about laptops Fri
> evening. It seems that my problems with th 2.4 kernel may be unique to
> my laptop. It's running a Cyrix MediaGX 180 MHz that I didn't think
> should be that different than any other sub-200MHz pentium, but it seems
> it is. Runs at 60 BogoMips instead of the 360 I would expect if it were
> a real pentium. This doesn't change the fact that the 2.4 kernel is not
> required for linux-wlan-ng. I currently have it running on the 2.2.16 RH
> kernel w/ pcmcia-3.1.19.
Ahh, but the MediaGX *is not* a pentium. I supports most i586 ops and
some i686 ops but is in no way the same CPU design. BogoMIPS are just
that BOGUS MIPS. They measure how many million times a second the CPU
runs a *specific* timing loop, and is used to calibrate a few things.
Bogos are comparable only between 2 of the exact same CPU (arch and
submodel), and only running the same kernel (I main exact same file
copied to two systems).
So if you had an Athlon XP 1700+ and a 2200+ both booting the same
kernel (single file copied to both systems) you could get a meaningful
comparison of their relative speed at running a small useless timing
loop. Any changes to this (different kernel versions, different
compile options, different compilers, diff CPU types) will throw the
comparison off, sometimes by a little, sometimes in a very big way.
(In the past there have been several major changes to the timing loop
code to better reflect what they are calibrating. If you compared say
an AMD 486DX4-133 booting 1.2.13 and 2.4 the bogos should be (IIRC)
about 2.5 times higher with 2.4 than 1.2.13. IIRC there were major
changes both in the 1.2.x -> 2.0.x and 2.0.x -> 2.2.x made to the
loop.
An abnormally low bogo rating is something to look into (for example
if you change a memory related option is BIOS and it drops 40 points,
look into it. Otherwise, ignore it. The MediaGX usually sees a
rating approx the same as its external bux speed (it has lousy
internals), for example a MediaGX 233Mhz STB I have shows 60something
Bogos with 2.4 (66Mhz CPU bus). This means almost nothing though, a
MediaGX 233 performs about like a Pentium MMX 180Mhz for real work
loads, it just sucks are running an extremely small code loop (it
takes a GX 3 - 4 clock cycles to run one loop, where and Athlon-XP
does a few optomizations and runs 2 complete loops *per* clock cycle),
and kernels/userspace almost never fall into this class.
--
I route, therefore you are.
---------
To unsubscribe, send email to <aklug-request@aklug.org>
with 'unsubscribe' in the message body.
This archive was generated by hypermail 2a23 : Mon Mar 03 2003 - 05:44:49 AKST