On Fri, 25 Jan 2008, Ferenc Wagner wrote:
"isplist@xxxxxxxxxxxx" <isplist@xxxxxxxxxxxx> writes:
What I was hoping to do with the partitions was to give each blade
it's own boot and scratch space, then allow each blade to have/use
shared storage for the rest. I was hoping to boot the machines from
FC or perhaps PXE.
I'm doing something like that on a BladeCenter. The (diskless) blades
boot using PXE/DHCP/TFTP from a central server. They can see over FC
two storage partitions. On the first, there is an LVM2 VG with a
single LV for the root filesystem of each blade. On the other there's
GFS which they all mount after booting up.
What about the overhead of managing unshared local roots? Nobody seems to
account for that, and IMO, it's signifficant.
This setup is simple, but lacks the shared root of OpenSharedroot. I
decided to go with the simpler design with more wasted space, because
the root volumes aren't particularly big, only a few GBs each.
Another refinement could be creating LVM snapshots of a common "image"
root LV for each booting server, which would again cut down on disk
space for the price of added complexity.
In another case, building several bigger but volatile clusters, I went
for a shared read-only NFS root with some symlinks into tmpfs's for
the files that had to be writeable. There was no SAN in that picture
at all, a read-write NFS share was enough, since locking wasn't an
issue.
Indeed, NFS can be almost as good as GFS if you don't need fully
operational locking. There are also scenarios where it's faster than GFS.
Gordan
--
Linux-cluster mailing list
Linux-cluster@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/linux-cluster
