[Gluster-users] Netboot / PXE-Boot from glusterfs?

Jan kirchy at gmx.de
Thu Jun 17 10:00:31 UTC 2010

Am 17.06.2010 11:37, schrieb Daniel Maher:
> On 06/17/2010 11:08 AM, Jan wrote:
>> - enables (diskless) Linux-Clients to boot over the network (debian,
>> some "real servers", some virtual ones)
>> Is this possible with glusterfs? Has anybody tried it? I understand it
>> is not that easy with a FUSE-filesystem, and there does not seem to be a
>> out-of-the-box solution, but has anybody tried it, maybe building custom
>> initrds?
> That's not what Gluster is for, really.  Fundamentally, Gluster is a 
> networked file system, and while it is highly malleable within that 
> context, it's not supposed to do anything other than provide a 
> networked file system.
> If you want to do network boots you'll need to look into another 
> solution to provide that functionality.  That said, in theory there's 
> nothing stopping you from building Gluster into your initialisation 
> image, then using it once your systems are booted and operational.

I could easily setup netboot the traditional way using NFS, but I would 
not have any failover/ha for that. As I understand, NFS on the gluster 
storage platform (gsp) does not provide failover in case the first 
server crashes.  Failover-functionality for some data being on glusterfs 
won't help when the root directory is "gone".

glusterfs sounds very promising, it seems to be a great fileserver for a 
bunch webservers and other shared/redundant data being used on many 
servers, it would make life much easier for me.

I (ideally) try to find a way to have such a shared file area, the 
netboot-server and ideally a few iscsi-luns on the same servers.

It is a kind of stupid to have a separate nfs-server just for serving a 
total of 1gb of files for some diskless servers to be able too boot 
(actually, it would have to be 2 servers again, with failover etc.. 
doesn't make sense that all diskless servers hang/crash when the nfs 
server crashes...). and I only need very few iscsi-luns with a total of 
10-15gb, not doing much io.

for NFS-netboot, I think there should be a way to use glusterfs instead. 
It's almost the same from the client point of view, the only problem ist 
the FUSE-client.

for an iscsi/drbd cluster, it is something totally different. I'm not 
sure if it makes sense to put that on the same server. If the gluster 
storage platform is a simple, stable solution that works out of the box, 
i'd probably prefer that and set up a seperate iscsi-cluster.


More information about the Gluster-users mailing list