[Linux-cluster] GFS / share storgage

[Jonathan E Brassow]

On Oct 25, 2004, at 11:11 AM, Michael Gale wrote:

> Hello,
>
> 	We are currently looking for a shared storage solution and have been 
> meeting with IBM about a SAN. All though a SAN is nice I want to be 
> able to offer a cheaper solution.
>
> I was looking at Apple's new Xserver RAID using ATA drives and I was 
> also reading up on GFS. I understand that neither may offer the speed 
> / performance of a SAN.

The Apple RAID has an FC external interface, right?  So, hooking that 
up to your fabric would provide you with a SAN.  I have no idea how 
their setup interface works -- perhaps web based.  Although these 
devices have two controllers, I don't think they are "active-active" - 
so don't think that that provides any redundancy.


> But does any body have any specs on GFS through put ?? or any ATA type 
> share storage device ?
>

I don't have any concrete data on throughput for your above questions.  
However, I don't see why ATA drives wouldn't perform well behind an FC 
controller.

GFS performance is tough to quantify.  Alot of it depends on the type 
of work load you have on your cluster.  When contention is low, GFS has 
been shown to scale near linearly with the number of machines.  When 
contention is high, performance suffers.  In general, you might use GFS 
in an HPC environment for performance reasons or you may use it for 
other applications for HA reasons.  Using a cluster file system wisely 
is like programming multi-threaded apps - just because you have 
multiple threads doesn't mean things will run faster - it's how you 
structure the work load.  The more intrinsic contention points you 
have, the more tuning will be required.

In the end, it may be wise to set-up a test environment and see if your 
envisioned solution performs as required.

  brassow