On Monday, 08/16/2010 at 07:27 EDT, "Bauer, Bobby (NIH/CIT) [E]" 
<[email protected]> wrote:
> We have been asked to make a zLinux server running under z/VM 100% 
> availability, except for POR (power on reset). We have 2 z/VM lpars, a 
test and 
> a production but as far as I know, z/VM doesn?t have a sysplex takeover 
> mechanism. 
> 
> I can move the server from one lpar to the other easy enough but that 
move 
> would disruptive in itself.

*Servers* are never 100% available.  A server is a single point of 
failure, whether it is a partition, a virtual machine, or a PC.  Things 
Happen.

*Applications*, on the other hand, can be made as close to 100% as is 
technologically possible because they can be designed to cooperate at a 
higher level, across multiple servers.  Therefore they can be designed to 
survive outage of a single server.  Can they survive the loss of two 
servers?  Only if there are more than two servers available.

z/VM having a SYSPLEX mechanism won't help you since SYSPLEX is an 
*application* service - it does nothing to improve the availability of an 
MVS address space.  [Yes, SYSPLEX is used by MVS subsystems, but they are 
just privileged applications in this context.]

What IBM *has* announced in a Statement of Direction is that in the future 
z/VM will have the ability to move a running virtual machine from one LPAR 
to another.  This improves the availability of the virtual machine 
dramatically for planned hw or z/VM outages, but it doesn't go to 100% 
since you can still have an unplanned outage (the network, the server, the 
LPAR, z/VM, or the guest). 

So, taken literally, the request cannot be achieved on any platform.  I 
can't imagine why They give a rat's behind whether a server is up or down. 
 Presumably They only care whether an *application* is available.  Solving 
application availability by making it a server problem is an architectural 
error of the highest order.  Making servers more available during planned 
outages helps to keep the *applications'* HA elements up and running. 
(Imagine if the International Space Station had only one coolant pump 
active at any one time.)

Alan Altmark
z/VM Development
IBM Endicott

Reply via email to