Hello,

some empirical results, checking how venus performs with a big cache.

I have filled a 5.3G venus cache with about 5G data.

top reports that venus has about 147M virtual image size (the rvm
data segment size is 558M, apparently lot of space is available (going
to be used for cml records ?))

The next access to all of the data causes no traffic.

The only problem with the setup is that venus does not fit in memory
(128 Mbyte RAM) and as it tends to regularly touch most of the rvm (?) the
other programs are being swapped out pretty consequently...

(what if we'd disable the hoarding daemon? In this situation I do not
need to hoard anything, all files fit in the cache)

It takes about 20 minutes to read 5 Gbyte cached data (~4.3 Mbyte/sec, not
too bad, considering a lot of small files) with 128 Mb RAM, and 14 minutes
the same operation (~6 Mbyte/sec, decent imho) if I add RAM so that the
same computer has 256 Mb. 466 MHz Celeron, Linux 2.4.23, ext3, Coda from
cvs 2004-02-27. IDE disk with 15 Mbyte/sec raw read speed.

A similar test on the underlying file system containing the venus
cache, e.g. almost the same data (a bit easier as the directories are
different) takes ~8 minutes, about 10.5 Mbyte/sec.

(possible use of the setup - an easy way to make a temporary offline
copy - walk through the tree, opening each file for reading, then discard
the token and disconnect the computer.... just ensure you are
Coda-disconnected when you want to use the host for data recovery -
otherwise you may inadvertently fetch updated files, or loose access to
the cached ones)

Regards,
--
Ivan