[gfarm-discuss:30199] Re: New to Gfarm: limitations?

[Osamu Tatebe <tatebe@xxxxxxxxxxxxxxxx>]

Hi Jordan,

> 1) Is there somewhere that I can find a list of the limitations of Gfarm? I
> have found a couple (like inability do a df or chown through fuse), but is
> there somewhere to get a complete list? The FAQ does not seem to focus on
> 2.0

Not for now.  Regarding gfarm2fs, the following functionalities
are not implemented;

- uid for another person
- gid
- nlink for directory
- readlink(2), symlink(2), link(2)
- chown(2) to another person (this is available by gfchown)
- statfs(2) (df does not work right now, but will be implemented soon)
- xattr

> 2) How stable of 2.0? Have people run it in large production environments,
> or is reliability and data safety in more of an early development stage?

We feel it is pretty stable.  We did several stress tests, and we
are now deploying in large production environments.

> 3) Is HA done automatically? For example, if I want each file to have 2
> copies in a 5 node cluster, and 1 node dies. Is a copy of the files from the
> dead node automatically copied to other machines so there is again 2 copies
> of each file? If I reinstall the dead node, will it then automatic
> rebuild/rebalance with files from various other nodes?

Basically it can be managed by an upper-level management layer
than the Gfarm file system, although we do not have such a system
fow now.   Regarding the reinstallation of the dead node, we
support the rebuild process by the -cc option of the gfsd.  

> 4) Is performance a focus of gfarm and has anyone used it as something like
> cluster temp space, or the focus more on highly available large volumes?

Instead of a cluster temp space, many people uses it for a
highly available large volume.

> 5) If anyone has any experience, how does Gfarm compare to some of the other
> distributed filesystems? Lustre lacks true HA (only through shared-storage
> and heartbeat), but GlustreFS and cleversafe look promising. Perhaps someone
> can shed some insight as to how Gfarm compares.

That is also we would like to do.

Thanks,
Osamu

On Tue, 8 Apr 2008 15:53:57 -0700
"Jordan Mendler" <jmendler@xxxxxxxx> wrote:

> Hi all,
> 
> I just installed gfarm on 1 node and will soon scale up it for our testing.
> We are looking at many options (including some of what I listed on
> http://jordan.spicylogic.com/blog/?p=8) for a backup system we are building
> to store lots and lots of genomic sequencing data. I have a couple questions
> though:
> 
> 
> 1) Is there somewhere that I can find a list of the limitations of Gfarm? I
> have found a couple (like inability do a df or chown through fuse), but is
> there somewhere to get a complete list? The FAQ does not seem to focus on
> 2.0
> 2) How stable of 2.0? Have people run it in large production environments,
> or is reliability and data safety in more of an early development stage?
> 3) Is HA done automatically? For example, if I want each file to have 2
> copies in a 5 node cluster, and 1 node dies. Is a copy of the files from the
> dead node automatically copied to other machines so there is again 2 copies
> of each file? If I reinstall the dead node, will it then automatic
> rebuild/rebalance with files from various other nodes?
> 4) Is performance a focus of gfarm and has anyone used it as something like
> cluster temp space, or the focus more on highly available large volumes?
> 5) If anyone has any experience, how does Gfarm compare to some of the other
> distributed filesystems? Lustre lacks true HA (only through shared-storage
> and heartbeat), but GlustreFS and cleversafe look promising. Perhaps someone
> can shed some insight as to how Gfarm compares.
> 
> 
> Thanks so much for the help. The fact that I was able to deploy an initial
> gfarm in a couple hours (including all prep and research) is very impressive
> and a sign of your good work.
> 
> Cordially,
> Jordan
>