Coda File System

Re: coda and RAID?

From: Jan Harkes <>
Date: Fri, 21 Jan 2005 11:52:46 -0500
On Wed, Jan 19, 2005 at 08:02:13PM +0500, Khalid Mehmood Khan wrote:
> Pardon me if I sound a bit irrelevant, but after skimming through
> coda's documentation and mailing list archives, I still find myself
> somewhat sketchy regarding coda's implementation and hardware
> requirements. I need to know:
> 1) Can we use some sort of a RAID for a coda server?  (silly I guess)

A Coda server is a normal userspace process, it stores metadata and file
data in regular files on the local filesystem. There is no reason why
RAID wouldn't work.

> 2) Can I use coda for cyrus-imapd mail-boxes and configdirectory?

Config directory definitely, I'm not sure about the mailboxes. Coda uses
optimistic replication without locking, so there will be conflicts. But
we only repair most types of directory conflicts automatically. So the
best way to store email in Coda is in maildir format. Several of my
mailboxes are stored in that format. If I am on a client that isn't
running Coda (which doesn't happen often), I access them remotely
through a courier-imap daemon which can handle maildir as the underlying

Another caveat with storing email in Coda in separate files is the
limited directory size. With the typical file name lengths of maildir
files we can only fit between 4000 and 5000 file names in a directory
before it hits the 256KB boundary. So my linux-kernel mail still has to
go to a folder on the local disk.

Finally, when the mail-delivery process loses it's authentication tokens
it cannot write to the email folders. For that I have my procmail
scripts set up so that any failed mail goes into a 'retry-delivery'
folder on the local disk. When I renew an expired token I run a script
that feeds all the emails in that folder back through procmail so that
they get redelivered in the right place.

> 3) What hardware would be sufficient ( good) for a coda server?

Not sure, our testserver ran for several years on a classic pentium 90
with only 64MB of memory or something like that. Ofcourse testserver
was only infrequently accessed. Until last summer our main production
servers were 200MHz PPro (or PII?) machines with 128-256MB of main
memory. During the summer we replaced all machines with 3GHz P4's with
gobs of memory. Comparitively they seem to be almost idle most of the
time, but startup times are great.

> 4) Is it necessary to have  identical storage on SCM and non-SCM? I
> mean if master is setup on a machine with three hard disks and
> completely or partially replicated to some other machine, do we need
> three hard disks on that machine too?

Not really. If you want to keep everything replicated you might want to
keep the amount of diskspace in the /vicepX partitions, which store the
file data, roughly the same.

Received on 2005-01-21 11:56:22