On Thu, 11 Dec 2003, Samir Patel wrote:

> Mainly for clean organization, but also for future scalability, I wanted
> to run two distinct servers... one would contain the applications and
> the other would contain user data.  Is it possible to do this using just

Ideally you would be able to migrate your data between servers
(and most probably you will be able at some point).

Technically it should be more efficient to distribute both types of data
across servers.

I a way, "servers" are totally independent of data semantics,
it is "volumes" and "paths" which would correspond to certain data types.

> one machine with just one NIC and one IP address?  The logical way of
> setting this up would be to use one large Coda server and appropriately
> organize the volumes.

Exactly. (or rather a couple of server, replication each other's data).

> But I'm not sure about Coda's upper bounds for
> server size.

Depends on your file size distribution, but roughly 30G file data for 1G
RVM (more space for RVM is hardly available on 32-bits machines)

> I might be confused, but the server installation scripts
> seems to suggest that using an RVM of 315MB is good enough for a almost
> 9GB of user data.  Is this true?

It is about the same estimation as the above.

>  What's the best way of creating a Coda
> server (or stack of servers on a single machine), that can hold
> somewhere between 40-60GB of data?

I'd go for either setting up a single server with as much RVM as I can
(1G), and hope that Coda will improve before I fill up the space :)
or a couple of 1G-RVM servers (two ip's on the same NIC is not a problem)
that should be quite enough for the data amount you mention.

>  The sizes of files will vary from
> just a few KB to up to a few MB.... might have some larger files, but
> very few >20MB files.

Sounds good for putting such data on Coda.

>  I know that people have discussed RVM limitations
> in the past, but I'm still not 100% clear on them.  Also, what's the
> easiest way to check RVM usage?

A bit unsure. I think it is not easy to guess how much space you have,
as RVM fragmentation can hinder (re)allocation of objects even when there
is some free space. I think it is worked upon, and hopefully newer rvm
allocation methods will allow for serving a lot more objects / more data
by a single server process...

>  (1) how does RVM size effect total server size and how can RVM be
> checked?

Old estimates sounded like rvm should be 4% of the data amount,
I think real rvm usage is less than that, as average file sizes
grew substantially since then. May be 2% will be fine nowadays? Don't
know.

Probably no way to surely predict how many objects can be added to the
server before rvm becomes full. Just hope for the best :) given the
rvm/data ratio you choose at the server setup.

>  (2) is it possible to run more than one server on one machine with a
> single IP address?

I don't think so, but let it have several ip addresses?

Take my suggestions with grain of salt, I did not test the limits, nor
multiserver-per-host setups.

Cheers,
--
Ivan