On Tue, Apr 02, 2002 at 11:47:45PM +0100, Jon Masters wrote:
> Hi,
>
> [apologies for lack of threading headers, I'll sort that out soon. ]
>
> Anyway, Matthew Sackman writes:
>
> > For those of you that use NFS and similar (or different), I've just
> > completed an investigation into stability and performance of ENBD
> > (similar to DRDB if you know that). If you're interested, the
> > report is available at http://www.namkas.com/enbd/
>
> An interesting report. I've not read it fully yet but a couple of points:
>
> * Check the graphs, more specifically the axis labels :-)
I know which one you mean. Lyx is taking the correct eps image and overlaying
another one on top of it. Lyx is not supposed to be able to do this and it's
very annoying. When I've fixed it I'll recreate the .dvi .pdf and .ps
> * Have you mailed this to Russel Coker, and if not may I?
> (he's the author of bonnie++ and would be interested in this).
Nope. If you would like to make him aware of this then I have no objection.
> * Have you tried this with *real* network cards? I use rtl cards myself
> but people will tend to moan unless 3com is mentioned somewhere :P
Um, the network that I was using contained 2 realteks, 1 3com 905CTXM and
1 Intel EtherExpress Pro 100.
> * You used software block devices, I reckon there's a significant
> performance issue there given the state of the loopback code (if that's
> in fact what you meant you were/are using?)
What I was telling the nbd-server was that the device to export is indeed
a file rather than a block device. However, enbd doesn't actually use the
loopback code (otherwise it basically wouldn't work). All things being
equal, I would agree that there is a performance penalty there. But I don't
actually think you'll see it until you get up to gigabit ethernet where
the network isn't the prime bottleneck.
> Anyway, a good read, I like (not that my opinion matters much). I'll go
> right through it later and see if I can add anything useful as I'm in the
> process of setting up distributed filesystems in a new cluster at home in
> Reading (the external facing side of which is apogee.jonmasters.org). At
> the moment it's going to be running NFS automounted filesystems with
> failover between machines and perhaps a distributed filesystem such as the
> (now non-Free :-() GFS or somesuch, but I am interested in exporting block
> devices and have played with this idea a little previously. I'm not sure
> of the practicalities of using it for a resilient RAID setup/system
> especially handling locking and recovering nodes and so forth - in fact I
> worry about the reliability of the idea but it'd be interesting to test it
> out and see what happens :-)
Well, my tests show ENBD is a hardy little sod. But then you don't get on
with 2.4 kernels and ENBD basically doesn't work on 2.2 kernels. With ENBD
you can only mount the device from one machine at a time. Therefore there
are no locking issues.
> The reason for going with NFS/NIS initially
> is that I'm setting up a playpen of Solairs 8 boxen at home which shall be
> joining the fun eventually...[0]
>
> Matthew, you coming tomorrow night?
Indeed. I was there. It was good to meet so many people IRL.
Matthew
--Matthew Sackman Nottingham England
BOFH Excuse Board: Cosmic ray particles crashed through the hard disk platter -------------------------------------------------------------------- http://www.lug.org.uk http://www.linuxportal.co.uk http://www.linuxjob.co.uk http://www.linuxshop.co.uk --------------------------------------------------------------------
This archive was generated by hypermail 2.1.3 : Thu 04 Apr 2002 - 23:32:44 BST