HA capabilities of ESX/ESXi Standard and Enterprize packages?

[stride]

Hello!

I'm considering various licenses for some inhouse solutions, HA being one of the options. Does any of you guys have any experience with this?

First of all, I have some questions when it comes to the HA and vMotion options in the Enterprice package. As I understand it gives you the possibility of moving VM's between different physucal servers, seamless to the user. Obviously this is good for planned maitenance etc. How does this work with unplanned outages - like some natural disaster (me?) suddenly powering off the box running an "active" VM?

When it comes to the "standard" package, I'm also curious of how HA is implemented, and what capabilities that is available.

For the solutions we are implementing, we also have an option of running "native" linux HA, for which the OS'es themself handle failover situations by moving IP's etc. Is the standard and enterprize packages worth the extra cost?

Any input would be greatly appreciated.

With best regards,

M.

[Texiwill]

Hello,

First of all, I have some questions when it comes to the HA and vMotion options in the Enterprice package. As I understand it gives you the possibility of moving VM's between different physucal servers, seamless to the user. Obviously this is good for planned maitenance etc. How does this work with unplanned outages - like some natural disaster (me?) suddenly powering off the box running an "active" VM?

VMotion only works if the host the VM is running on is powered on. Remeber HA is for disasters, VMotion is for balancing server loads, or for maintenance.

When it comes to the "standard" package, I'm also curious of how HA is implemented, and what capabilities that is available.

It is based on Legato AAM. If that helps.

For the solutions we are implementing, we also have an option of running "native" linux HA, for which the OS'es themself handle failover situations by moving IP's etc. Is the standard and enterprize packages worth the extra cost?

Well Linux HA is really a Linux CLuster and to do that you will have to setup a cluster per the the WP http://www.vmware.com/pdf/vi3_35/esx_3/vi3_35_25_u1_mscs.pdf while this WP is for MSCS, the Linux portion is really the creation of the virtual hardware (ignore the windows stuff). But you still need to follow these steps.

If you have VMware HA then Linux HA most likely is not necessary. You also can not use VMotion on a Clustered VM due to other constraints so in general you use either one or the other.

I have VMware HA and VMotion and never use a Linux HA cluster within my systems (I am about 80% Linux within my ESX hosts). I find there is no need to do so.

Best regards,
Edward L. Haletky
VMware Communities User Moderator
====
Author of the book 'VMWare ESX Server in the Enterprise: Planning and Securing Virtualization Servers', Copyright 2008 Pearson Education.
Blue Gears and SearchVMware Pro Blogs -- Top Virtualization Security Links -- Virtualization Security Round Table Podcast

[williambishop]

Is it worth it?

It's saved my butt several times. I know of people who it's saved as well. A lot of us run this thing in critical clinical environments....Patient care is affected if you're down, and you don't want that--EVER. A lot of others run it in critical banking, and other settings where seconds count. The ability to move a virtual machine to a newer, faster, groovier server, while it's still up and running? Priceless. The ability to perform a software upgrade and revert to snapshot instantly and online after a microsoft patch screws you for the fiftieth freaking time is just as priceless.

If I had it to do over again, I would. Every single time. I'd do it if it cost 10x more. In 2+ years, I've found that the most important ability in the stable is HA, second is DR. And the next version provides even more.

[mreferre]

To add on Ed's comments you might want to read this:

Just change the notion of "Microsoft Cluster Server" into "Linux HA".....

Massimo.

[mreferre]

Posting the link might help...

http://communities.vmware.com/message/1158732#1158732

Massimo.

[stride]

Ok... If I understand you correctly:

VMware HA, as available with the "standard licence", gives me an option of creating a single VM within a cluster of ESX installations of 2 or more servers. This VM will then have an active/running VM on one of the physical servers, while another is operataing in "slave" mode. If the physical box running the master fails, the slave will take over without any interruption?

vMotion, available in the enterprize package however, adds the flexibility of moving single running VM's which is not part of any HA-clusters around. This option is good for maintenance and automatic loadbalacing over a set of servers.

Please verify or correct me if I'm wrong

[rjtd]

Hi.

As far as I know, VMware HA works simply by turning on a VM machine on a different server when the primary server fails.

You can do the same thing manually, by using the "Migrate..." option on vCenter. VMware HA will just automate the thing.

On both situations you will see the machine off, and then powering on. Depending on the OS, it can take a while to boot up.

VMware HA operates by migrating the VM configuration data from one server to another. It does not have a hot "slave" server. It just powers on the machine on a different server.

Currently I'm using Linux-HA and I am more than happy. I'm using the iscsi initiators inside the VMs, instead of RDMs. It works perfectly, and it fails over much more quickly than having to boot up a VM from scratch. It also allows me to have active-active clusters, thing that VMware HA does not provide. But there are some things you must be aware, if you value your data:

• You will need shared storage

• OCFS2 is only supported on SUSE and Novell CS

• The only fencing method that is supported for VMs is the Novell sbd. Don't even try to use another method if you value your data

• You must be very carefull in enabling all the necessary watchdogs

• The failover machines must be powered on all the time (be sure to use active-active clusters to take some advantage of this).

Redhat CS will not work properly on VMs. There is no way for you to guarantee your data integrity, because they do not provide a totally reliable fencing mechanism for VMs (scripts with vCenter will screw up your data in some scenarios).

Be very carefull when you implement a cluster. There are dozens of scenarios that can screw your data, so be sure to use tried and supported methods.

[weinstein5 · Accepted solution]

VMware HA, as available with the "standard licence", gives me an option of creating a single VM within a cluster of ESX installations of 2 or more servers. This VM will then have an active/running VM on one of the physical servers, while another is operataing in "slave" mode. If the physical box running the master fails, the slave will take over without any interruption?

No - In VMware HA there is only a single VM - if the ESX Server or if the VM's heartbeat is lost it will restart on the other nodes in the HA Cluster - there is no slave VM that is shadowing - so yes there will be an interruption -

VMotion, available in the enterprize package however, adds the flexibility of moving single running VM's which is not part of any HA-clusters around. This option is good for maintenance and automatic loadbalacing over a set of servers.

Exactly - the automatic load balancing come witht the DRS option -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

[williambishop]

"No - In VMware HA thereis only a single VM - if the ESX Server or if the VM's heartbeat islost it will restart on the other nodes in the HA Cluster - there is noslave VM that is shadowing - so yes there will be an interruption - " Technically he's right...for today at least-- http://communities.vmware.com/message/1082860#1082860

[weinstein5]

True VMware Fault Toplerance is coming - http://www.vmware.com/products/fault-tolerance/ - but I only deal with the here and now -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

[stride]

Thanks all, I finally got it!

Your input and some online documentation got me on the right track. vmware_ha_wp.pdf (application/pdf-objekt)

The tiny note at the bottom of that document states that vm's with HA need to be stored on external shared storage. Obviously the hypervisors in the cluster do their own heartbeat thing, bringing up another vm based on what was last stored. VMware's term "restart" the service, which is used almost everywhere in their documentation now finally makes sence

[williambishop]

Don't forget to hit the video on the page weinstein linked to, it should probably be labeled TV-MA because it gets me excited every time! And it's just around the corner. Having used it, it's just as easy as they describe, and just as failure proof. The cool factor of things coming out with this next release can't even be listed...there's just too much.

FT = umm, umm good.