Coda File System

RAM/virtual/disk ratio and other questions

From: Anil B. Somayaji <>
Date: 16 Jan 1999 00:58:20 -0700

(This is a repost.  If you get two copies, sorry!)

Hello all!  I'm currently (as time permits) working on setting up a
coda server for our research group.  After doing some initial
experimenting on small machines, I've looked at the numbers for our
new server machine w.r.t. coda, and I got a bit worried.  Hopefully
somebody can tell me whether or not we need more hardware to balance
this machine out.

We've got a new machine all ready to be a server: dual PII-450, 384M
RAM, and two 18G SCSI disks.  Assuming that I want to use 30G for
coda, I believe I then need a 1228M RVM data partition.  Sure, I've
got plenty of disk space - no prob.  But does this mean I need 1.3G of
virtual memory for the server?  If so, still no prob, I've got disk
space.  But is this machine now going to thrash like a madman, with
_only_ 384M of RAM?  That's the big question for me!  I suspect this
is OS dependent, so FYI I was planning on running Debian 2.1 (slink)
on this box, with the latest 2.1/2.2 kernel.

Since I only have two disks, I was thinking of partitioning things
like this:

  disk 1: system, RVM log, swap, coda data
  disk 2: RVM data, coda data

Do I need to add a third disk (maybe to hold the log and swap
partitons) to get good performance?

I definitely want to do incremental dumps; initially, at least, I'm
only going to have one server running - so, is it ok to have
replicated volumes that aren't really replicated?  Can I change the
server replication list on-the-fly (to add or remove redundant
servers)?  Or do I need to create new replicated volumes, and transfer
data from one to the other?

I was thinking of running a kerberized coda setup.  We're currently
running NFS in our department, so it isn't as if things are really
secure at the moment; however, it would be nice to set things up to do
real authentication.  Is anyone else doing this?  Should I bother?
(We already have a kerberos server set up in our department (I'm not
sure which version), with its only duty being to authorize ppp access.
It would be nice if it could be used to do something more

Finally - in its current state (5.0.0), how does coda's performance
compare with NFS?  I know this is hard to measure, given the extremely
different semantics of each; but in practice, do users feel that coda
performance is on par, much better, or much worse than tuned NFS
installations?  Does that impression vary with network speed (i.e. for
10Mbit ethernet, coda wins, but for 100Mbit ethernet, NFS wins)?  I'm
sure that eventually coda should win, hands down; however, if I'm
going to get people here to try using an experimental coda server, it
would be good to know what sort of sales pitch I should give. :-)

That's it for now.  Thanks!

    (definite coda newbie)

- -- 
Anil Somayaji (
+1 505 872 3150

Version: 2.6.2

Received on 1999-01-18 12:46:02