GFS released under the GPL [LWN.net]

What GFS is:

Posted Jun 25, 2004 16:54 UTC (Fri) by stevenj (guest, #421) [Link] (3 responses)

GFS (Global File System) is a cluster file system. It allows a cluster of computers to simultaneously use a block device that is shared between them (with FC, iSCSI, NBD, etc...). GFS reads and writes to the block device like a local filesystem, but also uses a lock module to allow the computers coordinate their I/O so filesystem consistency is maintained. One of the nifty features of GFS is perfect consistency — changes made to the filesystem on one machine show up immediately on all other machines in the cluster.

In the future, it would be nice if the editors would summarize the relevant "who, what, why" information in their article; this isn't Slashdot, after all.

What GFS is:

Posted Jun 26, 2004 4:07 UTC (Sat) by djabsolut (guest, #12799) [Link] (2 responses)

obviously GFS differs from NFS, but what are the differences ?

What GFS is:

Posted Jun 26, 2004 5:03 UTC (Sat) by AnswerGuy (guest, #1256) [Link] (1 responses)

GFS is more of a SAN filesystem. The idea is that it can support concurrent access by multiple systems over a SCSI (multi-initiator), fibre channel or potentially even a firewire bus/fabric.

Of course one could then export this local/SAN fs over NFS to additional
clients. However, it's more interesting for custered applications that
need more tightly coupled access to their storage (where network and IP layers impose too much overhead --- latency in particular).

What GFS is:

Posted Jun 28, 2004 5:37 UTC (Mon) by daniel (guest, #3181) [Link]

GFS is more of a SAN filesystem. The idea is that it can support concurrent access by multiple systems over a SCSI (multi-initiator), fibre channel or potentially even a firewire bus/fabric.

Nice succinct answer. Additionally, GFS is also used with block devices exported over a network. For my own development, I don't have a shared disk at all, instead I export a SCSI drive using Hyperscsi. iSCSI will work as well. The grand GPL release also includes GNBD, a network block device exporter suitable for use with GFS.

Of course one could then export this local/SAN fs over NFS to additional clients.

That is one of several main applications of GFS out there in the field: basically, server amplification. You have a core of GFS nodes, each of which exports as many NFS clients as it can handle, and you can have a bigger NFS network for less money (because the cost of the cluster scales linearly, whereas the cost of a multicpu server of equivalent power scales quadradically).

You can also do html server amplification, particularly for sites generating dynamic content. A filesystem shared _efficiently_ between all the servers in an html farm is a big help both from the point of view of performance and maintainance.

However, it's more interesting for custered applications that need more tightly coupled access to their storage (where network and IP layers impose too much overhead --- latency in particular).

It's only interesting to applications that need tight coupling to their storage. Fortunately, that is most applications :-)

Network latency isn't much these days, particularly when you use something like HyperSCSI that sits right on the ethernet protocol, so doesn't incur the IP overhead. No TCP -> less latency. This is perfectly practical and useful.

Besides, GFS does a very good job of caching in-kernel, much as a local filesystem does, which tends to amortize the cost of network data transfers. And besides again, GigE gives Fiber Channel a pretty good run for its money.

GFS released under the GPL

Posted Jun 25, 2004 17:07 UTC (Fri) by smoogen (subscriber, #97) [Link]

From dealing with such things in the past, the delay in making the code GPL'd is usually due to the research that is done to make sure that all the code can be GPL'd. EG it didnt come from BSD/X11 code bases or that if it came from the outside that the other authors will give permission that it can be GPL'd. They also are probably needing to do 'background' checks on the code to make sure that a developer didnt take a short-cut and put in code they shouldnt have.

GFS released under the GPL

Posted Jun 25, 2004 20:05 UTC (Fri) by treed (guest, #11432) [Link] (3 responses)

GFS *used* to be GPL. Then Sistina (mistakenly IMHO) went proprietary with it. Since it was GPL before and all of the code came from Sistina it should have been easy to GPL it again. Unfortunately now we have OpenGFS and regular GFS to deal with. Hopefully they will merge again.

GFS released under the GPL

Posted Jun 28, 2004 5:50 UTC (Mon) by daniel (guest, #3181) [Link] (2 responses)

GFS *used* to be GPL. Then Sistina (mistakenly IMHO) went proprietary with it.

Indeed, _your_ opinion. I'm not so sure. GFS's tour of duty as a proprietary filesystem has helped it gain the fit and finish needed to work productively under commercial workloads. We're just very lucky that events fell out so that all that work came back to the community.

Unfortunately now we have OpenGFS and regular GFS to deal with. Hopefully they will merge again.

Speaking as an honorary member of the Open GFS team, we have already as much as merged. As we speak, Brian Jackson, our fearless leader, has prepared a GFS package for Gentoo, ready to be emerged. (I think somebody else even beat him to that, but you get the idea.) There will be more announcements on this and other interesting topics in the next few days, please stay tuned.

So, let's let the hacking begin. And oh, by the way, clusters are loads of fun.

GFS released under the GPL

Posted Jun 28, 2004 9:14 UTC (Mon) by mdekkers (guest, #85) [Link] (1 responses)

GFS's tour of duty as a proprietary filesystem has helped it gain the fit and finish needed to work productively under commercial workloads.

At the risk of sounding pedantic, are you suggesting that GPL / Open Source software does not usually have the "fit and finish to work productively under commercial workloads"? The manner in wich you pose the comment suggests that GFS would otherwise not be in such good shape.....

GFS released under the GPL

Posted Jun 28, 2004 15:17 UTC (Mon) by vmole (guest, #111) [Link]

No, he's suggesting that while it was proprietary, Sistina was able to devote more resources to developing it, because there was a direct cash stream associated with it.

Yes, there is a lot of good, reliable, free software. Some of it took a long time to get that way, because the developers simply could not devote the necessary timei to the project. Also, often the "final polish" part of a project, where you catch all the corner cases, is tedious and non-glamerous. It's lot more fun to start a new project, and since you're doing it for fun anyway, there's no incentive to get the last 1% right.

It's all well and good to be in favor of free software. I believe that in the end, you do have a better result, but it can take a while to get there, and ignoring that fact while rah-rah-rahing free software does no one any good.

at LAST

Posted Jun 26, 2004 1:54 UTC (Sat) by johnjones (guest, #5462) [Link]

at last !

a clusterFS

now if I could only get a MacOS and WinXP driver the world would be a better place
(the windows driver could cage off the openAFS stuff maybe maybe)

go on apple speak to redhat about it CUPS worked well in MacOS after all !

regards

John Jones

GFS released under the GPL

Posted Jun 26, 2004 21:16 UTC (Sat) by stevef (guest, #7712) [Link] (1 responses)

A detailed comparison of the Linux cluster filesystems on their technical merits would make a great technical paper (or Masters thesis). We are long overdue for a comparison of them on:
a) their network protocol, and how well the operations map to the Linux VFS
b) their Linux 2.6 implementation

I suspect that the most widely used on Linux are:
1) SANFS (Storage Tank)
2) GPFS
3) GFS
4) Lustre

But with the increase in ethernet speeds, the two popular network filesystems (perhaps with minor extensions) NFSv4 or CIFS, should be able to compete with SAN like approaches in some cluster environments as well.

A major problem with the cluster filesystems is that although some have open source implementations, they don't have standards documents written up for their metadata operations that sit above the block layer (iSCSI itself is pretty well documented but is lower layer). For the two most popular network filesystems they have deteailed protocol descriptions written up: obviously NFS in the IETF, and CIFS from SNIA.

The VFS improved a lot in 2.6 over 2.4 for network and cluster filesystems, with a few more changes to the VFS in 2.7 we may be able to achieve essentially local file i/o semantics over non-local filesystems which we can not quite do today.

GFS released under the GPL

Posted Jun 28, 2004 5:55 UTC (Mon) by daniel (guest, #3181) [Link]

The VFS improved a lot in 2.6 over 2.4 for network and cluster filesystems, with a few more changes to the VFS in 2.7 we may be able to achieve essentially local file i/o semantics over non-local filesystems which we can not quite do today.

Au contraire, GFS has already achieved that. The last of the GFS VFS patches (beyond a trivial export or two) is in play on lkml now, the cluster flock patch, which GFS doesn't need except to support local filesystem semantics for flock and fcntl locks.