inn 2.3.2 from isc.org compile problem -> --with-largefiles

Alex Kiernan alexk at demon.net
Mon Feb 4 12:35:13 UTC 2002


list-inn-workers at news.cistron.nl writes:

> *However* the history index and hash files will also get a lot
> larger because of the 64 bit file-offsets, so you will need more RAM
> to get the same performance. Especially as the VM in current (2.4.17)
> kernels isn't very good and likes to have lots of free RAM.
> 

I'm not sure anyones VM really knows how to deal with the access
patterns for the current history hash. The access is completely
random, so anything based on any kind of simple LRU model just ends up
getting it wrong much of the time - about the best I've managed is to
just lock the whole thing in core.

Also it degrades really badly when it spills from the first hash
table; I suspect that the pread/pwrite hammering we generate merely
serves to make the kernel cannibalise our carefully mapped first
table degrading performance still further.

-- 
Alex Kiernan, Principal Engineer, Development, Thus PLC


More information about the inn-workers mailing list