VMware Communities: Failover Clustering ...

VMware

Up to Discussions in VMware ESXi™ 4

This Question is Answered

1 "correct" answer available (10 pts) 1 "helpful" answer available (6 pts)

11 Replies Last post: Nov 9, 2009 9:21 AM by Mintofoxburr

Failover Clustering posted: Nov 6, 2009 7:20 AM

Mintofoxburr

3 posts since
Aug 20, 2009

This is more of a configuration question I am new to VMware so I apologies for my lack of knownledge. We have 4 ESX 4i Embedded servers connected to EqualLogic iSCSI SAN. I was hoping to build a couple of windows 2008 servers running Failover Clustering but it looks like from reading the VMware documentation this is not possible with the current configuration I am running which is 4 ESX Servers in a Cluster with DRS and HA enabled and also an iSCSI SAN as storage. Can anyone confirm this to be true and if so how could I do MSCS on VMware with my current configuration do I have to go back to physical servers for this to work?

Thanks in advance

Gary

Tags: mscs, failover, clustering

Re: Failover Clustering

1. Nov 6, 2009 8:07 AM

in response to: Mintofoxburr

DSTAVERT

2,388 posts since
Nov 30, 2003

You can certainly do MSCS with iSCSI directly inside the guest OS using the Microsoft iSCSI initiator.

http://www.microsoft.com/windowsserver2003/technologies/storage/iscsi/iscsicluster.mspx

Also http://kb.vmware.com/kb/1004617

Re: Failover Clustering

2. Nov 6, 2009 9:40 AM

in response to: Mintofoxburr

AndreTheGiant

5,897 posts since
Aug 28, 2008

See also: Clustering Software in a VMware environment

A failover cluster using iSCSI RDM disk is not supported by VMware... but it works.
Of you can use iSCSI inside the guest, in this case from VMware point of view you have two common VMs.

Andre

Re: Failover Clustering

3. Nov 6, 2009 2:07 PM

in response to: Mintofoxburr

COS

694 posts since
Dec 12, 2005

It sounds like you are trying to put a cluster (MSCS) on a cluster (ESX Ent Cluster). I highly recommend NOT doing this. Pick one clustering technology. If you are using DRS then you have your failover.

Re: Failover Clustering

4. Nov 6, 2009 4:05 PM

in response to: COS

EPL

56 posts since
Nov 5, 2005

Why wouldn't you want to run an MSCS cluster on an ESX Cluster? While ESX clustering provides some level of fault tolerance for that VM in case of hardware failure, it doesn't really solve the problem of availability of your OS. For instance, in the simplest terms, if you use a Windows 2008 Failover cluster for your file shares you can bring one of the members down for OS patches, reboots, etc, then fail it over to the other member once complete. This provides greater uptime for your services. Otherwise you're just installing your OS patches, reboot, then having your services available after the OS comes back online.

By the way, I'm running a windows 2008 cluster in vmware, running with Guest initiated iSCSI connections to my EQL SAN... haven't had any issues.

Re: Failover Clustering

5. Nov 6, 2009 4:51 PM

in response to: EPL

COS

694 posts since
Dec 12, 2005

So when DRS moves both your MSCS Nodes to the same ESX host and it (ESX Host) dies for some insane reason you're no better off than if you had just one Windows Server running in an ESX cluster, but you paid for 2 Windows licenses to achieve it. Sure having a cluster on a cluster is cool but is it really necessary or financially viable? Unless you work for a bank or large financial institution that requires that perceived high availability for transactional purposes then sure, it's a great solution. I've done the same as you 4yrs ago on an HP RX8640 using HP-UX 11i v2 with HP Integrity Virtual Machines running SQL Cluster Servers. No problems there. So I wouldn't think ESX would have any probs either. What i'm saying is having redundant redundancies is very redundant.

Re: Failover Clustering

6. Nov 6, 2009 5:16 PM

in response to: COS

EPL

56 posts since
Nov 5, 2005

I wouldn't allow DRS to place both nodes on a single host. There are rules you can set so that doesn't happen.

Unless you're using the FT features of vmware, which have their own set of limitations, vmware does little in the way of offering OS availability. You still have to restart your OS for patches and maintenance. What I'm trying to say is that a MSCS cluster takes care of that. If you have a business that is 24/7, and you need to have some of your services available to your users as much as possible, this is one way to do it. If you're a business that wants and/or needs this level of uptime and availability, then paying for an extra license of windows becomes moot. However if your organization allows for this sort of downtime, I can certainly conceed that its overkill. It all boils down to the business needs.

Re: Failover Clustering

7. Nov 9, 2009 8:35 AM

in response to: EPL

Mintofoxburr

3 posts since
Aug 20, 2009

Thanks EPL this is exactly the angle i am looking at as well my question would be did you have to dedicate a physical nic for the heartbeat connection?
I was looking at setting rules within DRS for each cluster server also just needed confirmation that this can be done. I also understand the argument of cluster inside a cluster but again look at one as a hardware cluster compared to a software based cluster for OS and patching.

Re: Failover Clustering

8. Nov 9, 2009 8:56 AM

in response to: Mintofoxburr

EPL

56 posts since
Nov 5, 2005

I didn't dedicate a seperate physical interface for the heartbeat, just created a new vlan and a new portgroup for that. One thing to note, in my testing, Windows 2008 server passes the Microsoft Cluster Validation test, however Windows 2008 R2 does not.

I haven't tested it lately, but for me R2 failed on detecting a Fibre Channel adapter, even though its all iSCSI (kind of weird). I was testing the R2 RTM, so maybe MS will release a patch for that.

Good luck

Re: Failover Clustering

9. Nov 9, 2009 9:13 AM

in response to: Mintofoxburr

COS

694 posts since
Dec 12, 2005

For best practices, you should use another NIC. EPL's method will work as well. The reason MS says to use a seperate NIC is, if your NIC load is high and your heartbeat is on it, you may get a hear beat timeout and a false failover may occur. I've seen this on SQL clusters in our Lab where a returning query that gets over a few million records saturates the physical NIC. Ideally though the heartbeat NIC should be on it's own segment, but we don't have the $$$ to keep asking for that.

Re: Failover Clustering

10. Nov 9, 2009 9:16 AM

in response to: EPL

COS

694 posts since
Dec 12, 2005

EPL....

"I haven't tested it lately, but for me R2 failed on detecting a Fibre Channel adapter, even though its all iSCSI (kind of weird). I was testing the R2 RTM, so maybe MS will release a patch for that."

Thanks for that bit of info. I haven't tried with iSCSI. I'll keep an eye out on this. Maybe I'll put a MSCS 2008 Cluster to test it. Thanks again. Did you try a fibre cluster by chance?

Re: Failover Clustering

11. Nov 9, 2009 9:21 AM

in response to: COS

Mintofoxburr

3 posts since
Aug 20, 2009

Thanks guys this is helpful I do have an available physical nic so I am going to test it with this method first just because of the other unsupported options and thanks about Win 2008 R2 i will try with just Windows 2008 first.

Go to original post

Search VMware Communities Advanced Search

Actions

View as PDF

Up to Discussions in VMware ESXi™ 4

This Question is Answered

Failover Clustering posted: Nov 6, 2009 7:20 AM

Re: Failover Clustering

Re: Failover Clustering

Re: Failover Clustering

Re: Failover Clustering

Re: Failover Clustering

Re: Failover Clustering

Re: Failover Clustering

Re: Failover Clustering

Re: Failover Clustering

Re: Failover Clustering

Re: Failover Clustering

Actions

Communities

Community Resources

VMworld

Get Support

Other Links

This Question is Answered

Failover Clustering posted: Nov 6, 2009 7:20 AM

Actions

VMware Developer

Developer Sample Code

VMworld Sessions & Labs

Purchase PSO Credits Online

Community Hardware Software

VMware vSphere

Community Resources

VMworld

Get Support

Other Links