[kwlug disc.] Benchmark discrepancies

[L.D. Paniak]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

> Is that because the routers do not have bzip2?
> 
> gzip is far gentler on the CPU compared to the CPU hog that is bzip2.
>  

The WRT54GL is running the 'White Russian' distribution which is
outdated and does not have a handy bzip2 package.  The 'kamikaze'
distribution with the 2.6.23 kernel has a much richer set of packages,
including bzip2.  I'm too lazy and fearful to mess with my perfectly
good home wireless system (WDS...) to upgrade the WRT54GL(s).  Maybe
someone with a WRT54GL running kamikaze would like to try out the bzip2
benchmark?

I agree bzip2 gives a much better workout.  Compare my earlier results
on the Asus with bzip2:

real	31m 10.15s
user	0m 0.16s
sys	11m 6.72s

A full 50% longer than gzip!


> Something is a bit odd here.
> 
> If you look at the user time, it is quite close (11 sec vs 6 sec).

Actually, 0.06s vs 0.11s -> almost zero!

> System time is not that far off either (11 min 8 sec vs. 11 min 29 secs).
> 
> But the sum of user and sys is off by A LOT when compared to real time
> (wall clock) (11 vs. 22 mins, and 11 vs. 19 mins).
> 

> So, there may be other factors here than raw CPU/RAM performance, e.g.
> I/O subsystem, random number generator?
>

> The sum of user and sys is close to real, meaning there is no wait for
> I/O or anything like that.
> 
> So, I don't know what is happening there.
> -- 

Here is a snapshot of 'top' on the router while the benchmark is running:

Mem: 16588K used, 13340K free, 0K shrd, 1356K buff, 7388K cached
CPU:  42% usr  57% sys   0% nice   0% idle   0% io   0% irq   0% softirq
Load average: 1.69 0.71 0.26
  PID  PPID USER     STAT   VSZ %MEM %CPU COMMAND
 3428  3426 root     R     2452   8%  57% dd if=/dev/urandom bs=32k
                  3427   3418 root     R     2652   9%  43% gzip -c
  3436 3418 root     R     2424   8%   0% top
  266     1 root     S     2436   8%   0% syslogd -C16
  255     1 root     S     2436   8%   0% /sbin/syslogd -C16 -S

It looks like everything is going according to expectations here: no io
waiting, no irq, the processor running 100%...

In the final time output, 'sys' returns almost exactly 57% of 'real' -
as it should.  It is 'user' that is out to lunch. For the Asus it should
be 8.19 min.

I think answer is that the time command (from Busybox) has some kind of
problem.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFH/ZtF8h2PnOHbiQcRAlevAJ41OAoHajmAf92jOu+csoB6fviygwCeKggA
yOVvxSs8ukilm0yfBjsnF7Y=
=MUVj
-----END PGP SIGNATURE-----