VMware Fault Tolerance is better than WFC

[Former Member]

Hi,

we currently have a 100% virtualized environment for SAP applications. Adding to that we have a Distributed environment where the Central Services are installed on a WFC setup, the application servers and central instance on a VM host. All on windows 2008 R2.

in my experience with WFC on VMware, it is highly complex and many areas which need to be carefully considered and configured. This needs a Support team with good skills and any incidents are difficult to troubleshoot due to the many moving parts. Due to this I have seen more failures on a WFC setup than I would have seen on a standalone Central Services/Gateway installation.

I am looking to see if a single host setup with VMware FT(fault Tolerance) will be a better setup I have read all papers and recommendations from VMware and SAP where they say why a WFC is a better solution but I am looking for a more practical solution. So far I only see the following downsides with a VMware FT based setup.

- 1 CPU limitation, yes that is a big one but if we install only the Central Services on the system, it should be fine. I have seen our WFC systems which has 2 CPU but I don't see a utilization of 10-15% ever.

- Rolling patching for OS patches - This is a big one, I am not sure how patching will work in a FT scenario, will I have to reboot the box which means downtime on the SAP Services, We usually use a simple failover for this and patch/reboot the systems one node at a time with limiting the downtime on the SAP application to the time it takes to failover the services to the other node. I have asked the question to VMware on how FT can be manipulated here but I think in the long run I can convince my business for a slightly more outage if it means a more stable SAP environment.

Looking for more practical suggestions and pre-cons to this approach.

regards

Yogesh

[former_member186495]

We recently migrated our SAP infrastructure from HP-UX to Windows, and been through this discussion.

it all depends on what your business requirements are and money you can invest.

Vm FT looks simple and easy to manage but you still have dependency on infrastructure layer to operate with the same understanding as application consultants.

WFC if configured correctly should be a seamless solution.

Vm FT will give you a downtime of more than what you have in WFC in failover/fail back  situation.

You will have no application monitoring in VM FT for any application issues.

In a VM failover situation you will loose all your inflight changes, compare to WFC.

some of these above points convinced us to opt for WFC.

-Kamal

[Rudi_Wiesmayr]

I disagree with this.

What you write sounds like you are mixing up HA and FT.

We just tested FT for a single-vCPU-ASCS and it did very well. The time for the switching to the FT secondary instance after killing the host the Primary ran on was NOT noticeable. This was REALLY seamless.

We currently have MSCS in our (old physical) production system and a failover of the Cluster Group where the central instance runs on IS a disruption which can only be avoided by running a pair of EnqueueReplicationServers.

This is much more complex to run and maintain than the very straightforward FT Approach.

So our decision was to leave the MS clustering behind us and move to a fully virtualized landscape with HA for all machines and - in Addition -  FT for the ASCS.

HTH, Rudi

[former_member186495]

Hi Rudi,

how would you manage the ERS in VM FT and changes/locks which are in midway when a failover happens.

-Kamal

[Rudi_Wiesmayr]

In the first line: NOT AT ALL.

This means: FT does a 1:1 mirroring of the running VM on another host. When the host fails, the secondary instance takes over immediately and the lock table is not lost ...

Second answer: I fell over a paper describing a FT-ASCS with a local ERS in it. Unfortunately it is only in my mind an I will have to find it again ...

Kind regards, Rudi

[Rudi_Wiesmayr]

http://scn.sap.com/docs/DOC-32023 says

Enqueue table resiliency with VMware Fault Tolerance
•enqueue table is protected against hardware failure
•enqueue table is not protected against OS failure
•with Enqueue Replication instance on the same OS, enqueue table is protected against enqueue server failure

[Former Member]

Hi Rudi,

thank you, all your comments are in line of what I thought the answer would be but I wanted someone who has implemented this to confirm.

I am not a big fan of WFC or MSCS as we used to call it. It causes more grief than benefit. It has too many moving parts and when using WFC in a virtualized environment, it becomes much more complex. We have over 20 SAP systems each with its own WFC for ASCS(SCS) and ERS on the nodes. It has been less than seamless for sure. it took us well over 2-3 months to become stable, get everything right as it was the first time, this is with VMware and Microsoft support. Now when we have few incidents with this, we have to upgrade to windows 2012. And guess what, SAP does not support in place upgrade of windows OS, mean we have to start all over again. I am not very keen to give it a try unless I have to and I am looking for another option.

Since you already have this place, my only question to you is how do you manage security and mandatory patching, something we have every month. Does  that mean you have an outage during every window, mean you take down the SAP application. In an WFC setup, we can patch node by node and avoid an outage on the SAP application. Users are ok with a slight fluctuation during a weekend. Not that it will be a showstopper but I would like to avoid an outage if at all possible.

regards

Yogesh

[Rudi_Wiesmayr]

We are very lucky: We have a weekly maintenance window of 30 minutes.

So you will envy us ...

Kind regards, Rudi

[former_member186495]

Hi Rudy,

Thanks foe the information, have you implemented and tried the above you mentioned?

Have a look at the last section of this forum

http://scn.sap.com/message/13557850#13557850

-Kamal

[Former Member]

Dang it. We have 2 hr monthly window I use for failovers and reboots of standalone instances. With HA and failover, don't have an outage on other applications. With FT this may change. I will see if I can figure something else out. thanks you all for your insights into this. 

[Former Member]

I would like to emphasize, FT is only solution if we use it for sap central services (SCS/ASCS), it has a single CPU limitation so it will not be able to support any other application components like DB, Dialog, even standalone gateway may not work.

Even with ASCS(SCS) it is recommended to do a full load test on 1 CPU before putting into production. The Memory limit is 64GB so it should be sufficient.

though I read somewhere that with Vsphere 6, it will support upto 4vCPU' but that may be Q2/Q3 2015.

regards

Yogesh

[Former Member]

>>And guess what, SAP does not support in place upgrade of windows OS,

>>mean we have to start all over again.

>>I am not very keen to give it a try unless I have to and I am looking for another option.

well, to be more precise:

Microsoft is not supporting an in place upgrade of the OS - at least not if ISV software is installed on.

So you have to uninstall all software on the box, then you can perform a in place upgrade of the OS, after that install the software again...

Also: in order to perform inplace upgrades of single WFC clusternodes you need to run a mixed version cluster. this is not supported at the moment.

Despite of Microsoft not supporting in place upgrades, SAP put a lot of efford in testing this. See also

regards

Peter

[Former Member]

Thanks for correcting, it is MS which does not support.

- in a VM environment, efforts to uninstall/reinstall applications and do an inplace upgrade are much more than building out a new system with an image. We have over 400 VM hosts, with an image, they can be built out in a matter of few days. Also with  parallel environment, we can stage a lot of work reducing overall downtime to a great extent.

thanks.

Yogesh

[Former Member]

yep,

if running VMware and planing for WFC guests you still have the problem that you need to use expensive storage for creating clustered shared disks (vSAN, iSCSI...). VHDs used by multiple VMs are still not yet supported??

In it's latest version (2012-R2) Hyper-V introduces generation 2 VMs with VHDx files, which can be used to build up guest clusters.

Still a problem with WFC-clusters: you can't clone them, at least not the existing once.

Microsoft has introduced AD-agnostic clusters in 2012 R2 which can be used to setup cloneable cluster configurations. But you can only clone a AD-agnostic cluster to another AD-agnostic cluster. It is not possible to clone non-AD-agnostic cluster.

FT still has some limitations on the network level (Mirror VM needs to be in the same subnet) which do not allow disaster tolerant configurations (more than 2 clusternodes in more than 2 datacenter, geographically distributed). It is not always a question of money...

regards

Peter

[Former Member]

yes, all technologies have their limitations, we need to figure out what is best of us and what can the business compromise. Also what can we afford, there are lot of other SAP supported cluster configurations but they are expensive.

One more question I have for FT replication if you have tried this. We use SRM to protect our VM's and replicate to the DR. can a FT VM be protected via SRM?

[Former Member]

sorry, I haven't.

maybe you should ask this question to VMware or a VMware related forum.

regards

Peter

[Rudi_Wiesmayr]

What do you mean by "WFC"?

I want to interprete your posting correctly ...

Kind regards, Rudi

[Former Member]

WFC is the common abbreviation for Windows Failover Cluster (configurations).

Most of us are still used to MSCS

kind regards

Peter

[Rudi_Wiesmayr]

Oh! Thanks!

Googling for "MSCS WFC" I found this: vSphere 5.5 Windows Failover Clustering Support

HTH, Rudi