[Info-vax] VMware

[Grant Taylor]

On 12/10/19 8:55 AM, Bob Gezelter wrote:
> Your assertion with regards to the instance being effectively quiesced 
> during migration is correct in concept, but it is incomplete.

Not necessarily.

> Large, if not the overwhelming majority of memory is not writeable or 
> not actively being written at any given instant. Present write and 
> I/O remain an issue. Block I/O initiation on mass storage and one 
> can quiesce those I/Os.. Network I/O will appear as lost packets,

I've done a number of migrations without loosing a single packet or 
dropping any connections.

Usually the worst is that it appears as if latency increased from low 
single digit ms to circa 10 ms for a packet or two before returning to 
the low single digit ms.

> recoverable by normal error recovery mechanisms.

I vary rarely see any hint that the guest OS or clients connected to it 
have any idea that anything happened.

> In effect, the situation is not particularly different than a 
> traditional power failure/recovery interrupt.

I think the situation is considerably different than a traditional power 
failure / recovery event.

Ad described in a previous response, VMware's HA feature means that the 
OS image continues running across multiple physical hosts.  Nothing has 
any idea that the original hardware is no longer being used.

Likewise with live migrations between hosts.  I can move guest VMs from 
host to host to host on any whim to my hearts content and neither the 
guest VM, nor any of it's connected clients, will have any idea that I'm 
playing games with it.



-- 
Grant. . . .
unix || die