[Gluster-devel] Using GlusterFS Instead Of GFS2

[Rick Rothstein]

Hi -

I'm building a cluster with pretty powerful server nodes, e.g.:

=== Dual Quad-Core Xeon's
=== Dual-Port Chelsio 10gb RDMA enabled NIC
=== Two 8-Port SAS raid controllers
=== 16 500GB Sata Drives,

and the folks assembling the hardware have suggested that I take a look at
using GlusterFS
in place of a somewhat "creative" use of GFS2
("creative" because neither MD-raid nor LVM2/CLVM-raid are cluster aware).

I really like what I'm seeing,

I'm planning to try giving ClusterFS a prototype try,

and based on reading some website info, I have a few questions:
---
On a single node, would it be OK to fire up 16 servers,
(one for each physical 500GB disk -- possibly using NUFA scheduling),
rather than one server for a gigantic 8TB, raid-6 disk?
---
How do GlusterFS "Mirror" Volumes sync-up after, say, reboot after a node
failure
(For MD-raid1, I place the Write-Intent-Log on a Solid State Disk)?
---
In the FAQ you discuss the bad points of Striping, but now that GlusterFS
provides a Striping capability, do you have usage guidelines.
(There are some areas, I might want to try using Striping to mitigate
usage_hot_spots)?

Thanks for any help.

Rick Rothstein