Coda File System

Re: Setting up a large file server

From: Ivan Popov <pin_at_medic.chalmers.se>
Date: Wed, 14 Sep 2005 14:39:39 +0200
Hi Jeremy,

On Wed, Sep 14, 2005 at 10:02:07AM +1200, Jeremy Bowen wrote:
> particular, I would like some advice about the RVM settings and how to 
> structure the volumes on my servers.

You want to set RVM as high as possible (I think the script offers 1G).

Volume allocation can vary widely, but as certain operations are
applicable on volume-basis only (like quota) or are serialized per volume,
thus reducing parallel throughput, you do not want big volumes.

I am allocating about 4 volumes per user, which is going to strike me
once because of limited number of volumes per server, but I hope that
limitation will be fixed some day, as well as the max number of files
per server.

> As I understand it, the /vicepa partition is where the actual user data is 
> contained. The RVM data and log partitions contain the meta-data associated 
> with the user data.
> 
> I'm concerned at the requirements for sizes of the RVM log and data partitions 

Do not use partitions, use files for RVM.

> as I'm going to be storing a large number of small files. The vice-setup 
> script indicates the RVM log partition/file should be 30M or less. Then the 

Right.

> RVM data size seems to max out at 1Gb which (as I understand) should handle 
> roughly 15-20Gb of user data.

Count with about 10^6 _files_ per server, the corresponding data amount
does not matter.

> Given that my existing file-server is handling an order of magnitude more 
> data, is Coda a suitable solution or should I be investigating alternative 
> distributed storage solutions ? 

It depends on how many files you have per server host.

   < 10^6 - you will be probably fine with a single Coda server per server host
 N * 10^6 - you will need at least N server processes with corresponding
            ip-numbers (!) on a server host

The above relates to each server host independently. It does not matter
in how many places each peace of data is replicated.

So, going over some number of files per server host is going to make
your life more complicated. Administration of one big chunk of a file space
is then replaced with administration of many chunks of limited capacity.

The decision whether you want to go that way is still yours.

Regards,
--
Ivan
Received on 2005-09-14 08:41:47