Coda File System

Re: Some questions...

From: Brian Jackson <brian-lists_at_mdrx.com>
Date: Wed, 25 Sep 2002 09:49:23 -0500
Sorry it took so long to reply I was trying to see how the answers panned 
out for you. 

If I understand you correctly Coda might be a little bit of overkill. If you 
just have 2 computers that are going to be using the data, you may want to 
try just a cluster{ed|ing} file system. OpenGFS, Oracle Cluster File System, 
or one of the others. There has been talk on those lists lately of using 
firewire as the shared disk, so no need for Fibre Channel. Don't know if 
this helps you at all. 

 --Brian Jackson 


Derek Simkowiak writes: 

> 
>      Hello,
> 	I want to use CODA for a load-balanced production webserver.  I
> plan on sharing my experience with the list once it's deployed.  But in
> the meantime, I need some help. 
> 
> 	I want to use CODA on a Linux Virtual Server-based, load-balanced
> cluster (using LVS-NAT).  This cluster will be using WebDAV to do
> read/write filesharing to multiple (~20 or so) clients, and general HTTP
> (read-only) website serving for a medium-sized website. 
> 
> 	I only need CODA for the purposes of replicating the hard drives
> safely acrossed the nodes of the cluster -- that is, the only "users" of
> CODA will be the nodes of the cluster itself.  My endusers will not have
> any CODA client installed.  Endusers will go to WindowsXP Explorer
> "Tools->Map Network Drive" and then type in the https:// URL for the
> load-balanced cluster.  They will drag and drop files to this WebDAV
> share, which will result in Apache/mod_dav writing a file to the the CODA
> filesystem, which will result in that file appearing on all of the nodes
> in the cluster, which will result in any future clients seeing that file
> regardless of which load-balanced node they actually get served by. 
> 
> 	CODA was chosen over NFS because of its instant replication
> service, and over Linux's "nbd.o" and "enbd.o" network block device
> modules (with RAID) because replication is atomic to files, not blocks,
> and because CODA can support more than one backup copy (i.e., several
> nodes can be in sync at once). 
> 
> 	Reading over the docs, some questions came up.  The first one
> is... what should the "big picture" of this setup look like?  Would each
> node in the cluster need to run both the CODA server _and_ client
> software?  I assume so, otherwise, how could the CODA server software
> replicate file changes to the rest of the servers, if those changes are
> not coming in via a CODA client?  But then... would each node be a client
> to itself?  Or would it be a client to the master CODA server (i.e., just
> one of the nodes chosen at random), requiring a config change/failover if
> the CODA master server crashes?  I really need an overview of how an
> experienced CODA admin would configure this. 
> 
> 	Next question: I want to serve about 80 Gigs of space via WebDAV
> (on the CODA filesystem).  Has CODA ever been used to serve up more than 2
> Gigs at a time?  The documentation says nothing about large volumes. 
> 
> 	Next, the HOWTO says to set aside 4% of your total volume space
> for RVM metadata storage.  For my setup, that would be 3.2 Gigs of RVM
> space for CODA.  But I want to use a file, not a raw partition (for
> several reasons).  Is using a huge 3.2 gig file for RVM metadata a
> plausable option?  Or should I just give up now and use the raw partition
> from the start?  (The docs say performance suffers when using a file, but
> does not give any indication of how much slower using a file is or if
> there are any limitations on the file size.) 
> 
> 	Next is about the Virtual Memory.  My servers each have 1 gig of
> RAM.  But according to the HOWTO, I would need 3.2 Gigs (4% of my 80 Gig
> share) of virtual memory _on top of_ any other applications I would need.
> I don't think Linux allows for a 3.2 Gig swap partition.  Will this be a
> showstopper for me?  Does this mean I can't serve up 80 Gigs of space? 
> 
> 	The next question comes from the HOWTO.  The HOWTO says "Do not go
> above 300 Meg on the [client-side] cache."  What is that 300 Meg limit?
> Is that simply an old number written down years ago?  Has any testing on
> the client-side cache size ever been done?  (Need more info, please.) 
> 
> 
> 	Basically, I want to make sure CODA is at least theoretically able
> to handle my needs, before I spend a bunch of time learning CODA, doing
> stress testing, etc.  We're happy to share whatever we learn with the list
> -- I'm sure the greater CODA community will benefit from the things we
> will learn from this project -- but in the meantime, any help is greatly
> appreciated. 
> 
> 
> Thank You,
> Derek Simkowiak
> dereks at itsite dot com 
> 
>  
> 
 
Received on 2002-09-25 10:49:16