2 x SANs 1 x Cluster - VMware failover (automated)

1

101001101

Weaksauce
Joined
Aug 8, 2013
Messages
69
Hi guys

I'm looking for some suggestions on a project I have set, basically I want to formulate my solution around the following:

2 x VMware ESXi 5.1 Hypervisors (currently I am battling between Dell and SuperMicro)
2 x SANs (leaning towards fiber, though I would love to go with Infiniband)
2 x FC switches
2 x Gigait switches (most likely C3750G's stacked)

The complexity here is two SAN's, the interconnectivity is straight forward enough with both hypervisors "seeing" all datastores; but I need to be able to failover automatically VM's between SAN's. Clearly there will be replication between SAN's handled by the SAN itself, but I need VMware to be able to bring failed over VM's online in the event that either one of the SANs is out of service.

I know I can use FT, HA cold boot, vsphere replication etc if I lose a Hypervisor, but what about where the VM exists entirely on datastore1 (SAN1) with SAN replication to SAN2 and SAN1 one is lost?

Site Recovery Manager looks a good fit for this but in all honesty I have never used it so I'm looking at you guys for some feedback based on field use, plus any other suggestions you might have.

FYI both SAN's are not geographically dispersed they are in the same DC location which helps.

On the subject of SuperMicro I have never used them in production / critical environments before, how do they compare? (I am not in the USA, so they are not as well known here but we have suppliers), any feedback (they are actually cheaper than Dell servers here) would be most welcome from production / critical environments.

SAN wise I am torn towards NetApp but previously I went with EMC (an ISCSI solution) - I have used both.

Now I am going to go and see if I can download a trial version of Site Recovery Manager :D

Thanks
 
The only way to automatically failover is to put a storage virtualization box in front of the two arrays..like an EMC VPLEX. Else it'll be a manual cutover.

Stick with FC... Infiniband isn't the answer here.
 
Another option, depending on the workload, is something like the HP Lefthand P4000 SANs. You can mirror the two arrays and if one fails the failover is transparent to vSphere. The downside is it's iSCSI.
 
Would Netapp cluster mode work here? I know they can protect against a controller failure but not sure if it would be able to transparently redirect LUN access to the second SAN in the cluster. But it seems like it should be possible if you have the LUN mirrored on both SANS.
 
The problem isn't mirroring..the problem is that the other SAN has to take the identity of the failed one. It's easy with iSCSI really...just an IP failover. But for FC? Initiators, targets, zoning, blah blah.
 
But don't you just add the LIFs on the cluster to the zone? That way it is not bound to the physical hardware.
 
kalthak said:
But don't you just add the LIFs on the cluster to the zone? That way it is not bound to the physical hardware.
Click to expand...

Very few SANs would support this. They just aren't built for it to do failover..much less automatic failover. They'd have to continually monitor the other array and only give you R/W access to the LUN if it failed. Not a simple task. Far easier for something in the middle like VPLEX.
 
What's your budget? For what you're trying to accomplish you're either going to have to use vSphere replication unless you go with NetApp or EMC and you're not going to get automated failover with SRM. If you go NetApp and you want automatic failover, then you're looking at MetroCluster. If you want EMC, then NetJunkie directed you down the right path with VPLEX. Both pretty costly solutions so make sure that's what you really want to accomplish and you want to pay for. If you can get away without automated failover, SRM is plenty good using vSphere Replication and if you go with NetApp for EMC, their SRA's work just fine as well. However NetApp requires SnapMirror for volume-based replication which is not free. I don't know about EMC, I don't use them. Again, you could still use vSphere Replication with either vendor arrays. Is automated failover a requirement or do you want it because it's 'cool'? Two hosts, this doesn't seem like a serious environment to warrant this type of expenditure. But I don't do your budgets :D
 
NetApp MetroCluster is not automatic, at least with SyncMirror anyway, you still need to initiate a controller takeover on the controller that holds the READ only LUN. There is a "tiebreaker" piece that you can add too this, i'm told it's really scripting, however I also think that even with this it's not fully automated...but could be wrong.

Lefthand would work as stated above with their "Network RAID" implementation and of course Distributed model as NetJunkie pointed out with VPLEX.

Correct me if i'm wrong but I believe you can put VPLEX "in front" of different vendor storage?
 
As QHalo said, can you give us some more info on what you are trying to accomplish?

I'm curious what the reasoning is behind the two SANs.
 
Vader said:
NetApp MetroCluster is not automatic, at least with SyncMirror anyway, you still need to initiate a controller takeover on the controller that holds the READ only LUN. There is a "tiebreaker" piece that you can add too this, i'm told it's really scripting, however I also think that even with this it's not fully automated...but could be wrong.

Lefthand would work as stated above with their "Network RAID" implementation and of course Distributed model as NetJunkie pointed out with VPLEX.

Correct me if i'm wrong but I believe you can put VPLEX "in front" of different vendor storage?
Click to expand...
It is depending on the condition. If it's a loss of controller, that's just an HA condition so automatic failover. If it's a disk failure and a controller failure then you have to manually do it. As long as the other controller can read the quorum on the dead controller's disks, automated failover can occur. If not, then you have to manually run 'forcetakeover' on the online one. This is how I understand how it works. It also requires 3200 and up controllers. Not cheap by any means.
 
QHalo said:
It is depending on the condition. If it's a loss of controller, that's just an HA condition so automatic failover. If it's a disk failure and a controller failure then you have to manually do it. As long as the other controller can read the quorum on the dead controller's disks, automated failover can occur. If not, then you have to manually run 'forcetakeover' on the online one. This is how I understand how it works. It also requires 3200 and up controllers. Not cheap by any means.
Click to expand...

We have a netapp 2040 and i'm in contact with the seller to buy second netapp and make a claster with them - he told me that it is easiest with FAS 3000 series and up - in case of 2xxx series it is duable but it will not be automatic.
 
This is cDOT(cluster mode vs 7-mode) not metro cluster. Two different things being talked about here.
 
shabazkilla said:
Another option, depending on the workload, is something like the HP Lefthand P4000 SANs. You can mirror the two arrays and if one fails the failover is transparent to vSphere. The downside is it's iSCSI.
Click to expand...

NetJunkie said:
The problem isn't mirroring..the problem is that the other SAN has to take the identity of the failed one. It's easy with iSCSI really...just an IP failover. But for FC? Initiators, targets, zoning, blah blah.
Click to expand...

This. It's how we're doing it with P4500 mirrored SAN's. It was the most painless solution for our needs.
 
vplex local will do it, and you can put dissimilar storage behind it, as long as the devices are the same size.

buuuut.. hope you got bank for vplex.
 
Hi guys

First off thank you very much for all of your replies, some great suggestions here and points to note - thanks again. I've been away for a few days so only just now got back.

You sure are right about VPLEX / MetroCluster pricing - it is astronomically expensive as I found out today, so this really leaves HP Lefthand P4x00 and StoreVirtual VSA(s).

There is no budget attached to this right now only a requirement, the requirement is to be able to absorb:

1. 50% loss of hardware in the solution, so this could be a switch, hypervisor or san with seamless failover / zero downtime

2. Where the SAN fails (tbh as you guys may concur, I have never had a SAN ever *completely* fail, thanks to dual everything - controllers, psu, ports etc but this is the requirement) and VMs are housed on the datastore(s) it is serving, failover needs to be with zero downtime (this rules out VMware HA, FT is also ruled out as it does not yet support multi-core VMs - pain).

3. The good - there is no geographic issues, the kit is in the same DC (I know this brings up political arguments around forking out for such a solution when its in the same room but....)

Having looked at this I am leaning towards HP VSA's and several 1U Dell servers with dual eight core procs (the spec went up over the past few days), Cisco 10GbE switches (4500-X-24's or Dell 8132's which are probably cheaper).

I have not used HP LeftHand VSA StoreVirtual in the field, only a lab, but it looks like I can have multiple P4x00 physical arrays and deploy VSA appliances to each ESXi hypervisor with datastores mirrored across the cluster and all accessed through a single virtual IP (iSCSI) created in Centralized Management Console via a Management Group. The area that worries me is the bandwidth required for the network RAID to function without latency, 10GbE is no doubt the MINIMUM but does anyone have any field experience with this in terms of network RAID lag / latency and how much data is being replicated?

Once again thanks guys for your input it is appreciated.
 
If you need zero downtime and VT restrictions are too onerous, have you looked at Stratus? Dunno if they're too pricey for you (full disclosure, I work there, although not in the vmware group...)
 
You say you have no budget, yet the solutions are expensive. Which is it? lol

Look for a solution that meets all the requirements first. Present that to them and then work from there. Pigeon-holing yourself into a half-assed solution is worse.
 
lol I thought someone might pickup on that when I read my post back, bad wording on my part - I think that with the price of MetroCluster and VPLEX it is going to break the bank (I had no idea it was so expensive); so the fallback will most likely end up being a single highly resilient SAN with multiple shelves (the default) but, having never used HP LeftHand and their VSA solution I see that as a middle ground, I've heard really good things about it but never had the opportunity to use it in the field.

My only concern is the bandwidth needed to perform functions such as network raid and latency / lag. Also latency with ISCSI compared to fibre channel. On paper it looks great and something I would love to put forward so I was hoping someone could chime in with their thoughts on that.

I've played with it (StoreVirtual VSA) in the lab using the 60 day trial and so far it looks very good, just waiting on a quote / price from our supplier. Not sure yet also which P4x00 array to combine it with, looking at options there tonight.

Thanks

EDIT: Just to add I'm not looking for someone to design a solution for me, that would be lazy, just appreciate any feedback on the options that are on the table in terms of field performance and application. :D
 
Unless you're talking about SSDs in your SAN you won't see any difference in latency between iSCSI and FC. Especially going 10.
 
You must log in or register to reply here.
Back
Top