Clustered VMs - High Availability vs Fault Tolerance

[Dean Powell]

We have a three node cluster and multiple virtual machines balanced across
the three nodes.

Most are running an application that requires a user initiated telnet
session. On those application service VM's I can Live Migrate and the
connections remain intact. When any other type of failover is attempted the
session is disconnected.

I have found myself in quite a debate regarding the differences between HA
and FT. I don't believe it would be possible to migrate the resource without
an impact to the telnet sessions, outside of live migration. Am I incorrect?

Should I expect to be able to configure a virtual machine in failover
clustering and expect no downtime? If so, can someone explain how that is
possible.

Thanks. Dean

[RCan]

Hi Dean,

I think you are correct, I would call a normal failover as "save state VM +
move LUN/ressource to other node and start it" process. The VM will be
paused here.

A LM is different here, in a LM process the VM will be running and the RAM
state will be migrated to other other node and during the last copy process
of RAM there will be a small timeout, normally (depends on network speed)
within a TCP timeout window. That means your telnet app should stay alive.

Hope that helps.

Regards
Ramazan

Post by Dean Powell
We have a three node cluster and multiple virtual machines balanced across
the three nodes.
Most are running an application that requires a user initiated telnet
session. On those application service VM's I can Live Migrate and the
connections remain intact. When any other type of failover is attempted the
session is disconnected.
I have found myself in quite a debate regarding the differences between HA
and FT. I don't believe it would be possible to migrate the resource without
an impact to the telnet sessions, outside of live migration. Am I incorrect?
Should I expect to be able to configure a virtual machine in failover
clustering and expect no downtime? If so, can someone explain how that is
possible.
Thanks. Dean

[Russ Kaufmann]

Post by Dean Powell
We have a three node cluster and multiple virtual machines balanced across
the three nodes.
Most are running an application that requires a user initiated telnet
session. On those application service VM's I can Live Migrate and the
connections remain intact. When any other type of failover is attempted the
session is disconnected.

This is the expected behaviour.

Post by Dean Powell
I have found myself in quite a debate regarding the differences between HA
and FT. I don't believe it would be possible to migrate the resource without
an impact to the telnet sessions, outside of live migration. Am I incorrect?

Failover clustering is HA. You are right that in HA, when there is a move or
a failover of a resource, clients will lose their connections and will have
to reconnect to the resource after the resource is started on another node.

Post by Dean Powell
Should I expect to be able to configure a virtual machine in failover
clustering and expect no downtime? If so, can someone explain how that is
possible.

In Live Migration, the resources (VMs) are moved with the memory on the
active location being copied live during the move process. For more info, I
recommend the following site:
http://www.microsoft.com/downloads/details.aspx?FamilyID=FDD083C6-3FC7-470B-8569-7E6A19FB0FDF&displaylang=en&displaylang=en

I hope it helps.