| [Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] |
> -----Original Message-----
> From: Greg Freemyer [mailto:freemyer-ml@NorcrossGroup.com]
> Sent: Thursday, July 31, 2003 1:05 AM
> To: opengfs-devel@lists.sourceforge.net
> Subject: RE: [ogfs-dev]Problem about md device.
>
> >
> > > The problem is if you want to run RAID in each of the
> ogfs nodes. I
> > > hope that is eventually supported by ogfs, but definately not yet.
> > >
> >
> > What would it take for OGFS to support this?
> >
> > -- Ben --
> >
>
> The Raid level code has to become cluster aware.
>
> In particular, if the raid level code does any caching, etc. then the
> caches have to have cluster-locks wrapped around them somehow.
I don't know of any raid level code within OGFS, but neither do I know that I've looked at all of the OGFS code ... is there any that you know of? If so, can you tell us where it is in the tree?
The model I've developed in my mind for OGFS is that OGFS (filesystem code, that is ... let's forget about pool in this discussion, if that's okay) just needs to be pointed to (mounted onto) a filesystem "device" (real or virtual). OGFS does not care how the filesystem device is made up (combining real devices to make virtual devices, RAID, whatever). It is up to the volume manager (or raid, or whatever) to put the filesystem device together, and OGFS filesystem code doesn't care how it is done; OGFS just needs to see a contiguous storage area presented on one "device".
So, I'm wondering exactly what you mean by RAID being "supported by" OGFS ... would there be anything that we would need to do within OGFS (filesystem) code to make it possible? Everything you wrote below here (MD stuff) seems to be outside of OGFS itself. I'm not clear about what you wrote above here ("raid level" code). I'm hoping that it, also, is outside of OGFS itself?
Bottom line, do we need to change anything *within* OGFS filesystem code to support (software) raid? (Sorry, late night redundancy!).
>
> Also any admin features of MD would need to be cluster aware.
>
> I have no idea how easy/difficult this is for MD.
>
> Stephen Dake, I believe, said he was working on this during
> some of our
> initial irc meetings.
>
> If that were done, then a very cost effective, yet reliable, solution
> should be buildable via Shared JBOD, Cluster-Aware MD, EVMS, and ogfs.
>
> (JBOD = Just a Bunch Of Disks (ie. no external raid
> intelligence, so low
> cost))
>
> Getting the above to be a reality would be fantastic from my
> perspective.
If the "don't care" OGFS model that I described is accurate, is Cluster-Aware MD the only missing link?
>
> Currently ogfs requires the shared storage to implement raid itself.
Not sure what you mean by this last sentence? I've been using OGFS with a 4-drive JBOD array; OGFS does not seem to require raid at all (although it does require some sort of volume or raid manager to put the drives together into a single filesystem device).
Are you saying simply that cluster-aware software raid is not currently available?
-- Ben --
Opinions are mine, not Intel's
>
> Greg
> --
> Greg Freemyer
-------------------------------------------------------
This SF.Net email sponsored by: Free pre-built ASP.NET sites including
Data Reports, E-commerce, Portals, and Forums are available now.
Download today and enter to win an XBOX or Visual Studio .NET.
http://aspnet.click-url.com/go/psa00100003ave/direct;at.aspnet_072303_01/01
_______________________________________________
Opengfs-devel mailing list
Opengfs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/opengfs-devel