Take the 2-minute tour ×
Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

I'm planning a redundant set-up for VMware. Two locations (about 8-12km distance) are to be connected with dark fibre for SAN connectivity. I am looking for a FC-SAN product that allows transparent failover for VMware Fault-Tolerance. This means, that if one SAN is offline the other SAN will promote the LUNs automatically .
Neither NetApp nor EMC does seem to provide this featue...

Any hint?

share|improve this question

6 Answers 6

What is transparent failover and do you really want it? Geo fault-tolerance is pretty difficult to do with any form of performance. Fault tolerance has already got a number of performance/scalability limitations, 1 vCPU, lock stepping etc, you need the same shared storage etc.

You do not in any circumstance want automatic DR failover, you want a big red button to push to execute it, coordinated in a formal decision making process. Small glitch and you are looking at a complex fail back process.

Both Netapp and EMC offer replication and storage failover in conjunction with VMware Site Recovery Manager. Netapp Snapmirror Sync or EMC SRDF but make sure you test performance of your apps etc before you enable this. If your sites are close enough together and you'd don't need to failover hosts etc you could look in to Netapp Metro Cluster with SyncMirror.

One way to do this would be if VMware added storage FT, writing a VMDK on two datastores.

share|improve this answer
    
Thanks for your input. I'm aware of the VMware limitations. However, I do not want a big red button; I want a fully automated failover system with manual recovery. –  CyberOptic Mar 22 '11 at 10:50
1  
I doubt you do :) But Netapp Metro Cluster could be an option if you really do. –  HampusLi Mar 22 '11 at 11:30
    
For VMware you would use it's Site Recovery Manager product which needs array integration for the LUN magic. I think one should probably point out that there should be no fault tolerant metro/geo storage available that should automagically promote the shadow LUNs from read-only to read/write and break the replication as this may be not nice to your application(s). All storage replication schemes/approaches work that way. –  pfo Mar 22 '11 at 15:33
    
pfo: AFAIK does the SRM not act automatically, meaning that the "big red button" has to be pushed? –  CyberOptic Mar 22 '11 at 15:37

HP EVA can do this with the right software. Specifically, you're looking for:

  • A pair of EVA6400/8400 disk arrays
  • Command View EVA
  • Continuous Access EVA
  • Cluster Extension Software

The second two bullet points are the meat of it. The last bullet point will be your second biggest cost after hardware.

There are some significant limitations to this, though. The replication between the two arrays has to be in Synchronous mode, so writes will only be considered committed if both arrays report it committed. 8-12km does introduce latency, and it could be very significant latency if your applications are sensitive to storage latency.

That said, it does what you're looking for. Automatic failover between disparate storage arrays without human intervention.


Unfortunately, the above is not VMWare-certified. If that matters to you, it's not doable.

share|improve this answer
    
Sounds like what I wanted, but it has to be certified... Thought the EVAs are certified already? –  CyberOptic Mar 22 '11 at 13:32
    
@cyberoptic they are, but the config is not. Vmware is picky that way. –  sysadmin1138 Mar 22 '11 at 14:34

HP StoreVirtual 4000 is the platform that does what you're looking on FC. HP P4000 family, formerly know as Lefthand, could do this years ago, on iSCSI.

HP StoreVirtual operating system (also known as Lefthand software, or SAN I/Q) allows distributed storage devices to be linked together into a network RAID pool. Depending on the network RAID you choose to use, AND in case the the storage devices are <3ms far on the network (e.g.: same rack, or two different sites with a fast connection), you can build a storage infrastructure that is physically in two different sites, while the storage infrastructure is considered as just one pool of storage devices: a single storage infrastructure with syncronous read-write capabilities. this means that if a site crashes, the other is online and available to the servers. Zero downtime.

While two HP EVA are two different storage devices (syncronized by an additional software layer), an HP StoreVirtual architecture is, by definition, a distributed network raid pool of n storage devices that continuously communicate one to each other. Building a redundant and n-"active" infrastructure.

The software part that does the magic is build into the HP StoreVirtual storage devices themselves and is the operating system itself (Lefthand operating system, that is SAN I/Q), so the High Availability and Disaster Recovery features of the storage system are "for free", included, into the cost of the hardware (storage). Wiht an EVA device, you need to buy the additional software that sincronizes two different units.

Vmware integration is 100% (application consistent storage snapshots are driven by the StoreVirtual devices that "stop" the VMware server infrastructure while they're executing them, for instance).

find infos on HP StoreVirtual products here: http://www.hp.com/go/storevirtual

StoreVirtual operating system software capabilities are available also into a virtual machine that you can load into your virtualization environment, called VSA (Virtual SAN Appliance). If you execute that virtual machine into your environment, ANY storage you already own (internal disks, as well) will be configurable as netowrk RAID HP Storevirtual disks and considered part of a distributed pool, with all the advantages of making remote copies, clones, snapshots, thin provisioned LUNs, ... that the lefthand operating system offers.

With the HP StoreVirtual VSA (Virtual SAN Appliance) you can transform whatsoever storage you own into a lefthand node. Of course, performances will NOT be certified on the hardware you own. High transactions environment will need to rely on a certified hardware architecture (the HP StoreVirtual devices themselves).

HP designed three different architectures called Virtualiazion Smart Bundles that you can use as working examples of virtualized infrastructures. One uses HP StoreVirtual VSA, to build HA architecture on VMware. Find it here: http://h18004.www1.hp.com/products/solutions/virtualization/virtvmware.html

VMWare certifies (please notice: it's VMware that certifies it) that the HP VirtualStore VSA support Vmware's metro cluster features on iSCSI. This really means, it's really working as expected (!).

:-)

share|improve this answer

What type of VM functionality are you looking at deploying, I do multi-site failover but I have the luxury of effectively static web and app VMs so just duplicate them around the place and use Cisco GSSs to load-balance over sites. Where I have problems is with my DBs which aren't VM'ed and are out of synchronous range so I have to use DataGuard/GoldenGate to async them. If your VMs aren't 'static/stateless' then you do need your LUNs replicated, have a look at HP and/or 3Par, they can do this in conjunction with VMWare's Site Recovery Manager but to be honest I'm surprised NetApp (whose boxes I despise btw) can't do what you want.

share|improve this answer
    
Thanks for your information. I want to utilize VMware FT. The guests and its applications/services are own-developments with an properitary TCP protocol stack. –  CyberOptic Mar 22 '11 at 13:31
    
That's going to be one seriously busy stetched FT vlan! –  Chopper3 Mar 22 '11 at 21:31
    
not that busy. i got about 3mbit max for 2x4 VMs w/ FT. and we've got a whole dark fibre to multiplex. –  CyberOptic Mar 23 '11 at 7:52
up vote 0 down vote accepted

FYI: I looked around and found DataCore SANsymphony that seems to provide the automatic failover functions. I'll have it deployed in our test environment.

share|improve this answer
    
Update after almost 4 years now. We rolled out the VMware FT and Datacore setup to production in 2011 and are running flawless since then. I did test the various failover states back then (hours of pulling chords, powering off servers/switches, documenting the behaviour) and have to say it works well. No "big red button" required :-P –  CyberOptic Dec 11 at 12:46

Maybe there is a Solution for your needs with Falconstor . Its a Storage Virtualisation that is able to transparently sync 2 FC Storage Systems certified with vmware . We have this Solution 2 years in Production and it always works nice.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.