Am Donnerstag, den 27.11.2008, 21:25 +0200 schrieb Pasi Kärkkäinen:
> On Thu, Nov 27, 2008 at 05:10:30PM +0100, Thomas Halinka wrote:
> > Am Samstag, den 22.11.2008, 20:19 +0400 schrieb Nicolas Ruiz:
> > > Hi all,
> > >
> > > I'm pretty interesting study this kind of solution for my office.
> > > Following questions from Pasi, i would like to know from you, Thomas,
> > > if you are using a SAN for your cluster.
> > i build up my own SAN with mdadm, lvm und vblade.
> > > If so, what kind of data access technologies you use with.
> > ATAoverEthernet, which sends ATA-commands over Ethernet (Layer2). It's
> > something like SAN over Ethernet and much faster than iscsi, since no
> > tcp/ip is used. also failover was very tricky with iscsi....
> > >
> > > Last question, how do you manage HA, Live migration and snapshots :
> > > owned scripts ?
> > heartbeat2 with crm and constraints and the rest is managed through
> > openqrm.
> Hmm.. so openqrm can take care of locking domU disks in dom0's for live
> migration? ie. making sure only single dom0 accesses domU disks at a time..
> > >
> > > This is interesting. Want to tell more about your setup? CLVM?
> > > iSCSI?
> > nope, just AoE and LVM
> > >
> > > What kind of physical server hardware? What kind of storage?
> > It s self-build. We had evaluated FC-SAN-Solutions, but they were slow,
> > unflexible and very expensive. We 're using Standard-Server with bonding
> > over 10Gbit-NICs
> > This setup transfers 1300 MB/s at the moment, is highly scaleable and
> > was about 70% cheaper than a FC-Solution.
> > >
> > > What exact kernel and Xen versions?
> > at the moment its xen 3.2 and 2.6.18-Kernel. I am evaluating 3.3 and
> > 2.6.26 atm.
> > >
> > > Thanks!
> > >
> > > -- Pasi
> > If interested in this Setup, i could get you a overview with a small
> > abstract, what is managed where and why... you know ;)
> Yeah.. picture would be nice :)
you can get it here: http://openqrm.com/storage-cluster.png
Some words to say:
- openqrm-server has mdadm started and sees all mdX-Devices
- openqrm-server knows a vg "data" with all mdX-Devices inside
- openqrm-server exports "lvol" to the LAN
- openqrm-server provides a boot-service (pxe), which: deploys a
XEN-Image to xen_1-X and puts this ressource into a puppet-class
in this xen-image is heartbeat2 with crm and constraints implemented.
puppet only alters the config for me...
- all the storage-boxes are standard-server with 4xGB-NICs, 24-SATA on
Areca Raid6 (areca is impressive, since write-performance of raid 5 =
raid 6 = raid 0). Only small OS and the rest of HDD is exported through
- header_a und b is heartbeat v1 cluster with drbd. drbd mirrors the
data for openqrm and heartbeat does HA for openqrm
- openqrm itself is the storage-header exporting all the data from the
storage-boxes to the clients
- openqrm-boot-service deploys a xen-image and puppet-configuration to
- all xen-server see all vblades and shelfes
- xen-vms resist on aoe-blades, so snapshotting, lvextend, resize2fs is
Storage: go buy 3 new server, put a bunch of harddisk inside, install
linux, install vblade and fire them. on the openqrm-server you only have
to create a new-md and extend the volume-group
Performance: buy a new Server and let him pxe-boot, create a new
appliance and watch your server rebooting, starting xen and
participating the cluster.
We Started the Cluster with about 110 GB-Storage - at the moment we have
about 430 GB Data and have to extend up to 1,2 PB in Summer 2009, which
is no problem.
No - go and search for a SAN-Solution like this and ask for the price ;)
i guess that we will be in summer in this performance-regions with about
30 % of the costs and much more flebility.
Price: Total: $ 1,635,434
> And thanks for the answer!
ay - cheers!
i will end up this post with some words of Coraid CEO Kemp:
"... We truly are a SAN solution, but SAN is not in the vocabulary of
Linux people, because SAN is equated with fiber channel, and fiber
channel is too expensive. But now, there's 'poor man SAN" 
> -- Pasi
Any Questions - ask me ;)
Xen-users mailing list