[ZODB-Dev] More timing data
Greg Ward
gward@mems-exchange.org
Wed, 11 Dec 2002 16:13:21 -0500
On 11 December 2002, Jeremy Hylton said:
> It's very difficult to measure process size in a meaningful way. Do
> you care what the peak memory utilization is? Or do you care about
> the working set size after a long time? Or something else?
<sarcasm degree="mild">
Actually, it's very easy to measure process size -- just depends on your
metric. If I can hear my hard disk constantly churning, and all my
windows move like treacle, and there's a l...o...n...g......
d...e...l...a...y for every keystroke, then some process is "too big".
Otherwise, not. For me, the "too big" process is almost always a ZODB
process -- the ZEO server, a ZEO client, or a FileStorage-opening
client.
</sarcasm>
> IIRC, when Jim and I measured the fsIndex() code we used a very large
> storage (only recall size > 1GB, not number of objects) and considered
> the RSS reported by top for a ZEO server. Don't recall what activity
> the server was performing. The memory utilization was about 4x bigger
> for the dict-based code.
Interesting... perhaps fsIndex only starts to pay off above a certain
size. Or there's something else in the trunk code causing memory
consumption to increase. I should try some tests w/ fsIndex disabled,
but I'm looking at other stuff right now.
Greg