In a zvm530 shop, last week we were testing this at a client.  We simulated a 
cable pull by forcibly DETaching a set RDEVs from the controller machine. Two 
great things happened:
1. it automatically recovered to the other set of devices (seemed to be like a 
second or so)
2. 1 minute later the DETached rdev came back as attached to the controller!

So there is a 1 minute timer loop in vswitch processing lurking, always lurking.

There are some PTFs in this area.
David

-----Original Message-----
From: The IBM z/VM Operating System on behalf of Mary Ellen Carollo
Sent: Tue 4/15/2008 11:01 AM
To: [email protected]
Subject: Re: [IBMVM] Workings of VSWITCH fail-over
 
Hans - 
The VSWITCH failover support attempts to initialize each of the devices =

defined for the virtual switch. If one attempt is successful the others =

are initialized as backups. 

In 5.2.0, if all devices are marked as broken manual intervention (SET =

VSWITCH CONNECT) is required to reconnect the virtual switch when the 
problem is fixed.  

The failover support was enhanced in 5.3.0. The VSWITCH periodically 
cycles between devices that report a "not ready" state (typically 
something like a cable pull) checking to see if one of the devices has =

been fixed and initialization can complete. 

Mary Ellen Carollo
z/VM Networking Development

On Mon, 14 Apr 2008 08:22:00 -0400, Hans Rempel <[EMAIL PROTECTED]>=
 
wrote:

>Our VSWITCH (z/VM 5.2) is defined with a backup OSA card going to a seco=
nd
>real switch. Both real switches were re-booted and we lost all access to=
 
VM
>system since the TCPIP stack is connect to that VSWITCH. During my 
testing I
>did see the fail-over work by using the backup OSA device but can't 
remember
>if the primary device re-connected as backup once OSA card became 
available
>again. I think z/VM 5.3 handles this better than z/VM 5.2.
>
>
>
>In any event the real switches came up again the real devices were ready=
 
and
>free. I had to issue a SET VSWITCH VSName RDEV addr1 addr2 from the HMC =
to
>acquire connection again. Is this as designed for z/VM 5.2 or a bug?  Is=

>there a better way? Can I define the VSWITCH to automatically re-attach =

the
>devices? I didn't see any other options to do this but then again I may =

have
>missed it. I hate to write a REXX program to kickoff once an address on =

the
>VSWITCH fails to reconnect that address.
>
>
>
>Hans
>
>
>
>


Reply via email to