Hi all

I have a customer who is thinking in vmotion (VMWARE vSphere) as database
failover solution instead of active/passive cluster or RAC.

I havent used this tool before, anyone know how does it work?

TIA

--
LSC

-- http://www.freelists.org/webpage/oracle-l

I hate to say it, but this is an off-topic post.
Still, I'll bite.
Its like the BOFH's boss' idea of storing the data in the network.
Not on network storage, but moving about in all of that GigE kit with the
blinkenlights.

During vMotion events, the VM is in 2 places at once - doesn't that imply
that an outage is twice as likely to occur?

What is the goal?

With RAC, multiple instances would remove loss of an instance as a single
point of failure.
So if there is no fail, no failover event would need to take place.
If a session is lost due to loss of an instance, a reconnection will find a
surviving node.
If that's not good enough ... head down the path of transparent application
failover with pre-spawned sessions.

VMware vSphere is a client side application for administration.
If your vSphere connection drops, simply reconnect.
The ESX/ESXi hypervisors keep right on chugging along whether you're
monitoring them or not.

vMotion can be used manually to rebalance loads to move a virtual machine
to a different node in the cluster, to relocate VMs prior to taking a node
down for maintenance or used automatically by the distributed resource
scheduler in VMware to do battle with any other load balancers that are in
place in the environment to obtain nice sinusoidal graphs.

By itself, vMotion does not provide failover.
This may be a good point to start - or you might want to start at a more
general level.
www.vmware.com/pdf/vsphere4/r41/vsp_41_availability.pdf

-bvmafh

-- http://www.freelists.org/webpage/oracle-l

We are using VMware (with VMotion) to provide HA for a cluster of Oracle databases and application servers running on Linux and it works very well, but Paul is right that it doesn't provide automatic failover.  This environment can tolerate the few minutes of downtime that would be required to restart the virtual host on another physical host and then have users reconnect.

I just skimmed the document Paul linked to below and apparently they offer a higher level of HA called "VMWare Fault Tolerance" that will provide faster, automated failover functionality but I don't have any experience with that.

Regards,
Brandon

Hi
Thanks for sgarng the experience.

So far I see two points with vMotion:

   1. *Hardware failure such as server brown out or network problems,
   automatic failover is possible*
   2. *Database/instance failure there is no automatic failover since
   VMWare does not detect application failures*

Am I correct?

TIA

--
LSC

-- http://www.freelists.org/webpage/oracle-l

I think that's almost right except that vMotion itself doesn't provide automatic failover - you need their HA product specifically for that and from my understanding they consider vMotion and HA to be two separate things although they may be bundled together depending on what package you buy.  The diagram at the link below shows that vMotion is included with "vSphere Standard", but you have to buy "vSphere Enterprise" to get their "High Availability" and "Fault Tolerance" products: http://www.vmware.com/products/vsphere/mid-size-and-enterpri[..]

Here is another link that might help clarify the differences:

http://communities.vmware.com/message/1864497

You'd probably be best speaking with one of their sales people (and have them include a tech resource too if possible) to get the details for your specific situation.

Regards,
Brandon

One unexpected "bonus" in this area is - if "Advanced" was licensed for
ESX/ESXi v4.x ... that can be converted to "Enterprise" for ESXi v5.x.
The downside is if the host in which ESXi is installed has 2 sockets but
say 96 GB of physical memory, the "Enterprise" license will only cover 32
GB vRAM per socket/processor license.
So one may gain features but have less memory to support the existing VMs
out of a "free" upgrade from "v4.x Advanced" to "v5.x Enterprise".
This could end up resulting in an additional "Enterprise" license being
required per pair of sockets in order to support the 96 GB of memory in
that host.
"Its a win win!" </sarcasm>

That's one way of dealing with ever-increasing processor core counts.

-- http://www.completestreets.org/faq.html http://safety.fhwa.dot.gov/ped_bike/docs/pamanual.pdf

-- http://www.freelists.org/webpage/oracle-l

oops. http://www.vmware.com/products/datacenter-virtualization/vsp[..]

The vRAM entitlement has been increased to 64 GB per socket for Enterprise
edition for ESXi v5.x.
"seasons greetings".
I'm sorry for posting stale, misinformation.

Additional bonus: SUSE Linux Enterprise Server for Linux.

I think that I know of some blades that might be getting a few DIMMs in
their respective stockings ...

Paul

-- http://www.completestreets.org/faq.html http://safety.fhwa.dot.gov/ped_bike/docs/pamanual.pdf

-- http://www.freelists.org/webpage/oracle-l