Hi,

Actually, we have a RHEL 4 with customized kernel for XFS fs support.
we must change the hardware of server and install a new server.

the server make PDC, file/print server (Samba) with ldap backend.

For prepare this migration, I look for a other filesystem than XFS because RH not support XFS and some times we have problem with some limitations of XFS (ex : 25 ACL entries) and also because we want to make this server in a high avaibility context.

For that I try to find a FS which permit to have a data replication over ethernet between to servers.
I know that GNBD or DRBD can make that, but each server will be connected to her SAN (2 servers, two SAN) with LVM.

Now, I want to know what is the best FS or system which permit to have data mirroring between to server and support LVM. The configuration can be master -> slave or master <-> master.

I looked GFS and GNBD but the GNBD and lvm compatibility is not sure and I don't want have problem
with data corruption.

thanks

There was some days ago a thread on the gfs-mailing list that nobody yet tried GFS with gnbd and data replication.
But the outcome was that it might work with gndb and cmirror as part of the devicemapper.

But neither ways I would be extremely concerned about performance and synchronisations of i/os.

When using DRBD you end up in the same dilema as before. IMHO DRBD is not offically part of RHEL4/5 and therefor not supportet. Besides you would need DRBD/GNBD/cLVM/GFS for GFS or DRBD/ext3 for ext3. I won't use it with GFS.

BTW: Basically this is independent from the filesystem. So you are more or less free to use any filesystem on top.

Advantage for GFS is that it is a clusterfilesystem that you can concurrently mount and access on all nodes belonging to a cluster and build up a active/active Samba cluster. So no double mounts or stuck unmounts when doing the failover.

I would extremly advice you to do a lot of testing and think about the possibly of other ways to do the replication (or leave it because of complexity and performance) or use the storagesystem functionality if available.

Have fun trying
MG.

Thanks for your answer.

More informations : I want to a HA servers.
My servers will be HP DL servers and OS = RHEL4 (RHEL5 is too young).

The data will be stocked on SAN and I make a LVM layer over SAN DISK and my problem is :
what filesystem can be choice with these functionnality

- POSIX ACL support.
- Performance (same as XFS or more)
- supported without problem by samba.
- large number of inode.
- Easy to configure and use.
I'm not sure that I must set LVM for SAN disk, it's a rule with my hardware vendor.
Anyone have experience or benchmark with GFS or GPFS ?

thanks

I agree to not use RHEL5 yet.

GFS does support all the features you are requesting:

• Posix ACLs
• supported by samba without problems
• large number of inodes (what is large and what kind of access do you have??). This is with a cluster filesystem a crucial point. Especially with GFS.
• Easy to configure and use: If you are already using XFS you'll find GFS also quite easy. But you are going to have a cluster and that brings complexities!!
• Performance: Is always a point. When you compare XFS with GFS as lokal filesystem - because that is the only fair comparison - you won't find to many differences. But you'll plan to use a clusterfilesytem as it is and then you cannot expect it to be scaling linear over the node. The point are the types of i/os.

Last but not least use LVM cause it brings in loads of easyness for storagemanagement.

GPFS vs GFS:

I'm not the experianced with GPFS but IMHO it is
1. quite expensive
2. is tied to IBM Hardware
3. has its strengths with big big filesystems (hundreds of TB and more) with big files and loads of nodes.
4. GPFS is not symmetric as GFS is. That means there are nodes for special tasks like Metadata handling and Locking. That might slow down performance when loads of filesystem i/os are used.
5. Looks like ACLs are only supported on AIX

But I'm an expert on GPFS. I only know some basics.

Have fun
MG.