2

There is a Xen cluster that setup that I really like. It consists of 2 nodes (let's call them 1 and 2). Each node has 2 non-raided drives (A and B). A1 replicates to A2 through DRBD and B2 to B1.
When machine 1 goes down, VMs come up on machine 2 and vice versa (just with 1/2 the performance)

I'm happy with this. It's cheap, flexible and robust, but it's not perfect ... and it's not letting me rest. Maybe you can help me make it better (or have an alternative architecture that's better than this ... I'm not against VMWare or other if it does the job)

Here are my problems:

  • Xen doesn't have a convenient converter like VMWare does. Or does it? (Disclaimer: I'm in love with VMWare converter)
  • DRBD is still unknown (scary even?) for most IT guys. I do not want to be the only expert that knows how to work this thing. I want things as dumbed down as VMware makes them. Some kind of easy web frontend for DRBD maybe? Even better, is there some kind of an appliance that does this (in the style of freeNAS, monowall and friends) Or some kind of a plug and play product? I don't mind paying more if it means that I'm no longer rolling my own and get some kind of vendor support.
  • Related to the above problem: Last time I've had a system like this was a few years ago, and I always felt like I'm on bleeding edge, and playing with things too much. It was a lot of fun. It didn't have the rock solid enterprise feel to it as VMWare does. This time around I don't want any "fun" :-) ... I just want it to work and never stop working. Maybe that was because I was still on Xen 2.0 and DRBD 0.7, I'm sure a lot has change since then, but something tells me that I'm still in for a lot of custom kernel compiles and challenging glitches now and again. Something I want to avoid at all costs this time around.
  • I was running this across a single 1Gb ethernet. What's the best way to double or triple that that keeping the above points in mind?

... maybe I could have summed all of the above with. How do I dumb down and fool proof this Xen cluster architecture?

Thank you

Tomas
  • 39
  • 2
  • If you don't want to be the only one that knows it about DRBD, did you document your setup thoroughly and start training someone in your department to get him or her up to speed? – Bart Silverstrim Dec 09 '11 at 14:20
  • I don't want to break your balls here, but asking for the moon on a stick is somewhat matching here.. There is usually some cost involved to get the things you want without having to hack it together. – pauska Dec 09 '11 at 14:21
  • As I said, I realize there is a cost, and I may be willing to pay it (if there is a product to do this). I'm still baffled why Citrix XenServer support this out of the box. I'd be more than happy to hand some extra cash for DRBD wrapped in one easy and robust package. – Tomas Dec 09 '11 at 14:34
  • 1
    Because it's a complicated set of technologies, all done with different projects, and you're trying to find someone willing to keep up with the releases of each version of each project and make sure they all work together (heartbeat, DRBD, VM management). If XenServer supports it out of box, you could license that and use it instead...that's why it tends to cost more. – Bart Silverstrim Dec 09 '11 at 14:39
  • Usually by the time a company needs something like clustering and it gets to a size where you *need* management of a higher degree, you have a large company's solution if you don't roll your own. So you MacGuyver it as you're doing now, or you pay through the nose for an integrated solution. Smaller projects that try something like this are so niche that the payoff is negligible and there's little thanks to keep the project moving, so it becomes outdated/abandoned. – Bart Silverstrim Dec 09 '11 at 14:41
  • I would agree with what you are saying as "the current state of the insudstry" .. however I would not call this a niche requirement. Everyone needs reliability, even a small office. I think of this 2 node cluster setup as RAID1 for entire servers ... don't understand why this isn't as wide spread as RAID1 ... this is not a 10,000 node cluster we are talking about here. – Tomas Dec 09 '11 at 20:32

4 Answers4

1

I don't know if you can dumb it down and foolproof it completely. There are management programs that can help aggregate and manage various clustering technology, but for your specific setup you're still going to have things that are unique to your setup.

You'll also have issues because you're tying various technologies of different versions together. You can get something that'll manage Xen, but Xen plus DRBD may take more work, or other software may not stay up to date with various releases.

Your best bet is to thoroughly document your configuration, complete with diagrams, and train someone to the point where if you're hit by a bus tomorrow that person can take over for you.

And whenever something is altered on that cluster in any way, update your documentation and make sure your team knows about it.

Bart Silverstrim
  • 31,172
  • 9
  • 67
  • 87
1
  • We all love VMWare Converter, it's useful even if you don't use VMWare virtualization products (I've used it to migrate different Windows physical servers to VMWare player, and converted them to KVM, it will probably works for Xen too);
  • I am not using DRBD anymore, I prefer using distributed fault-tolerant network filesystems like GlusterFS and MooseFS, you can achieve better scalability for your data than a ~RAID1 like DRBD.
  • You don't need to stay on the bleeding edge, but you need to stay update, your customers will appreciate it.
  • Upgrading to 10gb or switch to NAS/SAN for storage (Infiniband, fibrechannel, etc.)
Giovanni Toraldo
  • 2,587
  • 19
  • 27
0

I found that XEN, DRBD and Heartbeat is a very good combination. DRBD delivers Heartbeat-Scripts for its devices, DRBD delivers XEN integration by adding a device-type "drbd".

If you want all in "one hand" - take SLES11 (current patchlevel - SP2 currently) with HA-extension - then all three components will have support. DRBD has a partnership with SLES, so it should work without problems.

We are currently in the progress of upgrading our server-network to 10 GB so we can use live-migration (XEN) und faster DRBD relplication.

In our setup a XEN PV DomU has on "disk" consisting of a DRBD device, which consists of a LV. We use two-node-clusters with Heartbeat V1 Syntax (with the older SLES10 - we had to compile drbd 8 and had to write our own xen-heartbeat-script).

We are using XEN for its strenghts regarding Linux PV.

Nils
  • 7,695
  • 3
  • 34
  • 73
0

There actually are DRBD web based management consoles...

http://www.drbd.org/mc/management-console/ http://copilotco.com/mail-archives/webmin.2009/msg00486.html

dtbnguyen
  • 322
  • 1
  • 6