Information required for Geoclustering ESXi5

[tairatcliff]

Hi guys,

I've been asked to compile a presentation to my employer in regards to the future (next year) of their virtual environment, including but not limited to created a cluster of their ESX hosts across 2 Datacenters.

I've been looking around for as much information as I can get on the pro's and con's of setting up an active/active cluster across data centers that a in different geographical locations.

Can anyone point me in the direction of good information?

Here are a couple points...

The link beetween data centers is 10 gigabit (can possibly have multiple links)

Will have around 250 VMs across 14 hosts

This will be approached as a replacement for SRM

EMC FC SAN replicated at both datacenters

There are a couple things that i'm attempting to get out of this..

facts that are for and against getting rid of SRM

The work involved in setting up both the SAN and ESXi cluster in a production environment

If vCenter/ESXi5 offers a great oppertunity to geocluster than vCenter4.1/ESX4.1... I read that vCenter 5 offers a single vSwitch across an entire cluster??

How HA will/won't work in a geocluster

If FT can be implemented across a geocluster

+ any other points anyone can reccomend taking into consideration.

Personally i'm happy with the Active/Passive that is currently in place and utilising SRM in the even of a Disaster Recovery. The only driving force behind this is currently the cost that will be involved in renewing the SRM licences to a per VM basis and the fact that half of the current hosts at the Secondary Data Center are not being utilised.

any opinions are appreciated.

[Josh26]

Hi,

I'm not aware of there being anything new in version 5.

If you were to ask me about a geographically dispersed cluster, SAN replication and SRM tend to be how you do it "properly".

The closest option you have to an active/active cluster is something like a Lefthand mirror - whilst accepting the fact that for one site, all storage traffic is going to go down the WAN. It's really not an ideal environment.

You've been able to create a distributed switch that spans all hosts in your cluster since before version 5.

[keiooz]

I've been using ESXi5 for quite some time, but haven't been able to notice the part where Geoclustering is involved. If you were to tell me, I haven't seen any upgrades that will focus on this area, but I am still not sure of it.

[Virtualinfra]

tairatcliff wrote:

Hi guys,

The link beetween data centers is 10 gigabit (can possibly have multiple links)

Will have around 250 VMs across 14 hosts

This will be approached as a replacement for SRM

EMC FC SAN replicated at both datacenters

facts that are for and against getting rid of SRM

The work involved in setting up both the SAN and ESXi cluster in a production environment

If vCenter/ESXi5 offers a great oppertunity to geocluster than vCenter4.1/ESX4.1... I read that vCenter 5 offers a single vSwitch across an entire cluster??

How HA will/won't work in a geocluster

If FT can be implemented across a geocluster

Geo cluster is a new solution for DR.

1. With Geo cluster you can use both the site as Active\Active.

2. Less down Time.

3. HA works in geocluster, but you need to VM host affinity rule to RUN VM with the site A or site B, so only during site failover it will move the VMs to site B

4. FT can be implemented across a geocluster, but its not recommended personally speaking because of the distance.

5. Vsphere 5 provides better improvement with HA options, so its better to use vsphere 5 since architecture change is in place.

6. 10 Gig connectivity between site with <5ms latence to be noted,, but if you use enterprise plus then it can be less than <10 ms, since the enterprise plus is capable of with standing <10 ms latency..

refer the below KB of testing geocluster.

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=102121...

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=102669...

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=200754...

[Lessi001]

Hi,

in vSphere 5 there is a new Feature which will help you in a geocluster - it is called Metro vMotion (take a look at yellow-bricks http://www.yellow-bricks.com/2011/08/03/vsphere-5-metro-vmotion/). This feature allows you to safely vMotion a virtual machine across a link of up to 10 miliseconds RTT instead of 5 ms.

Maybe also take a look at http://www.yellow-bricks.com/2011/10/05/vsphere-5-0-ha-and-metro-stretched-cluster-solutions/.

Regards

Andi

[vMario156]

Which kind of EMC SAN you are using?

To realize a real streched cluster you will need a layer between a active / passiv storage system (like Clarrion/Celerra/VNX with MirrorView/Replicator) to get an active/active system. On EMC side its the VPLEX solutions as far as I know, where you of course need min. 2 systems.

Even if I don´t know the pricing I guess you can buy a lot of SRM licences for this money

But may you allready have such kind solution??

Regards,

Mario

[depping]

In order to do this your array needs to be capable of dealing with this type of set up. In many cases that means EMC VPLEX. Now note that this is not the same as an SRM solution, things that make your life easy in SRM like workflow orchestration / scripts / testing... all that is gone with a stretched solution. Links to some of my articles on how to do it are already posted, but I would discuss internally first what the requirements are of the solution. Stretched is not full DR solution like SRM offers.

[Virtualinfra]

Just curious any reason for why its not  full DR solution?..

Your comments would be appericated..

[Josh26]

Reading the KB articles posted, specifically this part:

On creation of a VPLEX distributed volume, a winner VPLEX cluster must be assigned. In the event of VPLEX inter-cluster communication failure, the winner VPLEX cluster continues to service I/Os destined to the VPLEX distributed volume. The loser VPLEX cluster does not service I/Os received on the distributed volume.

It looks just like an EMC version of this:

The closest option you have to an active/active cluster is something like a Lefthand mirror - whilst accepting the fact that for one site, all storage traffic is going to go down the WAN.

Have I understood this right?

[depping]

With a DR solution I expect to be able to recover from failures in an orchestrated way, defining things like startup order etc. I expect to be able to test my DR scenario without any disruption to the workload it self. A stretched environment does not really offer you this today. Yes you can define "HA boot order", but this is no guarantee whatsoever that VMs will actually boot in that order.

I like stretched cluster concepts, don't get me wrong, but architecting one is not as easy as it sounds.

[depping]

Also note that 10MS is only for vMotion latency. In a Geo Cluster 5MS is still a requirement for your storage replication network.

Moving this thread to the Availability sub forum where it belongs...