As Rosenblum shows with VMotion, the application gets a "warm HA' blanket as it moves to another server:
Rosenblum started a server on stage that was running the equivalent of 50 users pounding on Microsoft Outlook. The server's ongoing activity was being mirrored on a second server, which was receiving a live stream of events as they were entered into the log of the virtual machine on the first server.
As Rosenblum unplugged the first server, VMware's management software, Virtual Infrastructure 3, detected a failure and shifted handling the users to the secondary server. Since the secondary server was already receiving a stream of log events, it could pick up at the precise point where the other had left off. The pause between one virtual machine stopping and the secondary server's virtual machine starting appeared to be about a second.
but coming back to the applications:
- Scalability: What if they had 20,000 users in outlook? will it still take a second?
- Clustering within VI infrastructure: Would it be better to clusters the applications, like with marathon's everRun (love the name!), to have a continuous HA, which is more of the "Hot HA"
- Optimization: Using tools like that of Inovawave, we will have to make our applications perform even better
- Application itself: can it be developed to be SMP and multi-core aware?
- 64 bit: Do applications perform better as the 64 bit bandwagon apparently seems to purport?
- Compatibility or customization for load balanced applications: Will all user sessions be transferred to the other ESX servers? Or will the applications, say for instance Blackboard (a mission critical application for Academic institutions) which is currently using a rather cumbersome load balancing mechanism, will perform better with a single 64bit instance?