Hi all,
I've done some testing between the ac22 flavours.
Both kernels have identical characteristics (same .config, same reiserfs
3.6.9, same modified pegasus driver).
Testing 'methodology' as follows:
* boot
* let GDM start and settle
* log in
* let the stuff settle (at which point, a GNOME environment with two
xvt's is loader)
* in one of the xterms, cd tmp/testing
* uname -a >>$(WHICH)one
* ./do_ps >>$(WHICH)one # ps with fancy args, script included.
* vmstat 1 >>$(WHICH)one
while vmstat runs,
* launch netscape
* load two pages on a local Zope+Postgresql 'server' combination.
Those page include quite a lot of stuff and mildly complex queries (but the
results volume *is* big. Pedagogic referentials can be quite big <grin/>)
* switch to the second xterm
* cd $SOMEWHERE
* emacs main.tex & rm *.dvi
* in emacs, modify a letter, keep emacs around
* lambda main.tex ; odvips -o main.ps main.dvi
* switch back to netscape. Reload the first page. Load the second
page on a different record.
* switch back to the first xterm. Stop the vmstat
* ./do_ps >>$(WHICH)one
This looks to me quite real world, at least, it's almost exactly one of the
tasks this machine will have to do (edit and compile latex files while
keeping an eye on that *#$! ref.).
Results (meminfo, cpuinfo, and raw results attached below) :
firstone.gz: (ac22-riel)
299 lines
15 Mb max cache used
20 Mb max swap used
1.3 Mb max buffers used
secone.gz: (ac22-class)
300 lines
13.5 Mb max cache used
18.8 Mb max swap used
1.2 Mb max buffers used
(out of lazyness, the "megabytes" are half-baked between real Mb and hard
drive manufacturers Mb, that is, 1024000 bytes).
Basically, Classzone stole a bit less of RAM for the cache (still too much,
IMHO: my processes are still too much swapped out, overfavourising the
kernel's 'needs' (I want my RAM back. Less cache, please.)). (on final, emacs ended up with
twice as much RSS with classzone than without. Zope wasn't that lucky ;
netscape and postgres were pretty much unaffected). I'd tend to say, based
on swap an cache results, that classzone does a slightly better job on this
machine ; however, the wall clock time was roughly the same (vmstat's output
size used as a measure of time).
On a subjective side, ac22-class feels a little bit snappier on the long
run, but it seems to start apps (and wake them out of swap) slower than
ac22-riel. Result is, I don't know which one to choose (last booted -class,
will stick to it for the moment).
Baseline machine is a Toshiba 230CDX, with a P133 (non-MMX), 32 Mb of RAM,
1.5Gb reiserfs /, 8 Mb ext2 /boot, Debian potato as of yesterday, and much
room for optimisations.
-- Cyrille
-- Grumpf.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Jun 23 2000 - 21:00:20 EST