userdb cache expiry

[Tim Dijkstra]

On Mon, 18 Sep 2006 16:05:08 -0400
Havoc Pennington <hp at redhat.com> wrote:

> Tim Dijkstra wrote:
> > 
> > Why doesn't dbus just rely on the OS to do the caching? On linux,
> > nscd exist for this purpose for example ...
> > 
> 
> It was just too slow, I forget why but there were a *lot* of lookups. 
> Maybe the real issue was just when running the test suite.
> 
> This is something that should be fixed, I agree. As a first pass you 
> could try hacking the cache so it never caches anything and see what 
> difference it makes in time to run "make check" - probably better to 
> develop a simple real-world benchmark, though.

FWIW, I commented out the part that gets the user from the cache
(AFAICS there is only one function that directly access the cache). Not
the part that puts it in, because that would create a memleak, which
would abort the test (I didn't feel like investigating it further). So
the test is even a bit unfair, in addition to looking up the user the
nocache case also tries to add the user all time. 
Despite all this there doesn't seem to be significant difference
between the to cases; 

With cache

real    1m36.253s
user    0m17.533s
sys     0m6.684s

real    1m39.846s
user    0m17.945s
sys     0m7.036s

real    1m27.192s
user    0m17.685s
sys     0m6.864s

real    1m15.803s
user    0m17.053s
sys     0m6.692s

real    1m11.619s
user    0m18.337s
sys     0m7.272s


No Cache

real    1m53.428s
user    0m19.329s
sys     0m9.225s

real    1m55.809s
user    0m20.713s
sys     0m8.449s

real    1m16.380s
user    0m18.481s
sys     0m7.364s

real    1m25.100s
user    0m18.393s
sys     0m7.668s

real    1m11.619s
user    0m18.337s
sys     0m7.272s

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.freedesktop.org/archives/dbus/attachments/20060920/e8c14b26/signature.pgp