On Wed, 2002-08-07 at 07:20, Martin J. Bligh wrote:
> A few observations.
> 1. According to sar you're mostly idle (well, the machine is ;-))
> Both on 2.4 and 2.5 ... so if you're IO bound, kernel profiles
> are not going to be as fascinating.
Yea, with the single disk for I/O and the rather small memory
you run into that. But I would expect the numbers to be
uniformly low. (The one way 2.4 kernel seems low so ....)
> 2. Seems like your mem is full of pagecache, even before you
> start. I presume from the stats & the lowmem you're paging, or
> at the very least getting into page replacement which might be
> sending you into pathalogical hell ... would be very interesting
> to compare with 2.5.26 (ie without rmap), and see if that makes
> any difference.
I agree that it is in some sort of pathological hell (aren't we
all in one way or another?) The question of course is how did
are level in hell change between OS versions?
> 3. What kind of IO (raw?) are you doing to what kind of filesystem
> or not-a-filesystem?
> 4. BR, total 948.446274, db 948.416236, server_db_time 948.44625
> > I believe this means _all_ the query time was spent some where inside
> > the db engine...
> > It just ran __really _sloooooow with bunches of system time.
> Me confused ... you said it's all time in the DB, then you said
> bunches of system time - is this just a typo and you mean user time?
> 5. For 2.4, looks like you might want Bill's statm_pgd_range patch.
Timothy D. Witham - Lab Director - wookie@...
Open Source Development Lab Inc - A non-profit corporation
15275 SW Koll Parkway - Suite H - Beaverton OR, 97006
(503)-626-2455 x11 (office) (503)-702-2871 (cell)