[Linux-cluster] iSCSI GFS

[Balagopal Pillai]

On Sat, 26 Jan 2008, Johannes Russek wrote:

Hi,

     I have tried http://iscsitarget.sourceforge.net/ a few times and 
it was quite easy to setup in the simplest of configurations. The 
performance was comparable to gnbd. But gnbd has the extra advantage of
having a fencing option available for using with gfs, while with iscsi 
you have to rely on ipmi,drac,ilo etc etc. With a sas das for use with 
gfs, gnbd is a good choice. I tried this almost an year ago when i was 
evaluating an alternate cluster file system. The performance was nice, 
but gnbd had stability issues on very high load. So went back to Lustre
and happy with the choice. I used the iscsi enterprise target before to 
export a big storage volume from the internal net to be mounted on an nfs 
server and later exported to the public net via nfs. It was stable for my 
use. There is also the new FCOE coming soon to provide FC over ethernet. 
That is a also choice for someone who has invested so much on FC, but 
wants to take advantage of cheaper ethernet without losing the reliability 
of fc. Might need dedicated switch hardware though! 
http://www.open-fcoe.org/ But the cheapest solution for those people with 
FC is still re-export via a software target like ietd. Just my thoughts!

Regards
Balagopal



 

> Hi Mike,
> 
> > In fact, it got me wondering if my FC path might not be the greatest based
> > on Gordan's (I think it was) input on the subject. I've also seen a few
> > posts about iSCSI lately. Can anyone shed a little light on iSCSI and what I
> > would need to give it a try.
> >   
> I've heard that http://www.openfiler.com/ is probably the easiest way to get
> an iSCSI target up and running. Didn't try it yet, though.
> But i've got to say, that SAN and performance thing really depends on how you
> do things. I've got a five node Cluster with GFS over FC running here, and it
> is pretty fast. Compared to friends of mine who use SAS-DAS, the bulk
> performance was about as fast. We are mostly serving files from it (FTP/HTTP),
> so write-performance is not so much an issue. It might be that it's usually
> slower then local disks, but I guess it really depends on your overbooking
> ratio. We are happy with the performance :)
> Also I wonder if iSCSI does protect Block-Integrity as FC does. That's one of
> the things that make FC so robust, isn't it?
> Anyone knows anything about this?
> enjoy,
> Johannes
> 
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster
>