Hipersocket routing choices question

[Brian Nielsen]

If I understand correctly, by definition I can't use a VSWITCH to share
a hipersocket connection the way an OSA is shared. This leaves the choice
of using a TCPIP stack as a router or connecting each guest directly to
the hipersocket.

I've read the performance report that compares direct OSA connections
with routing through a TCPIP stack and with sharing the OSA with a
VSWITCH, and I'm not sure how/if those results can be extrapolated to the
hipersocket connection choices above. While I know the direct hipersocket
connection will/should be faster than routing through a TCPIP stack I was
wondering if anyone can point me to some more relevent performance
references.

We have about a dozen LINUX guests that connect to a guest LAN and
route traffic through a TCPIP stack to a z/OS LPAR. From a configuration
management standpoint, is there any reason I should consider keeping the
route through the TCPIP stack rather than giving each guest a direct
connection to the hipersocket?

From DR standpoint it seems to be a wash since we'll recovering under a
vendors running VM system.

Brian Nielsen

[Rich Smrcina]

I have a similar situation, except that VSE is in the other LPAR. For
Linux machines that need to communicate with VSE I assign them a
hipersocket interface. If it's a situation where performance isn't a
big concern (an occasional communication requirement), I just let the
Vswitch handle it.

Post by Brian Nielsen
If I understand correctly, by definition I can't use a VSWITCH to share
a hipersocket connection the way an OSA is shared. This leaves the choice
of using a TCPIP stack as a router or connecting each guest directly to
the hipersocket.
I've read the performance report that compares direct OSA connections
with routing through a TCPIP stack and with sharing the OSA with a
VSWITCH, and I'm not sure how/if those results can be extrapolated to the
hipersocket connection choices above. While I know the direct hipersocket
connection will/should be faster than routing through a TCPIP stack I was
wondering if anyone can point me to some more relevent performance
references.
We have about a dozen LINUX guests that connect to a guest LAN and
route traffic through a TCPIP stack to a z/OS LPAR. From a configuration
management standpoint, is there any reason I should consider keeping the
route through the TCPIP stack rather than giving each guest a direct
connection to the hipersocket?
From DR standpoint it seems to be a wash since we'll recovering under a
vendors running VM system.
Brian Nielsen

--
Rich Smrcina
VM Assist, Inc.
Main: [phone]
Cell: [phone]
Ans Service: [phone]
rich.smrcina at vmassist.com

Catch the WAVV! http://www.wavv.org
WAVV 2006 - Chattanooga, TN - April 7-11, 2006

[Brian Nielsen]

- Subnet A is the external network extended by VSWITCH to reach the Linux
guests
- Subnet B is the internal HiperSocket network connecting z/OS and the
Linux guests
Alan Altmark
z/VM Development
IBM Endicott
=========================================================================

So would I. I just wanted to make sure there wasn't anything I overlooked
in favor of keeping the routing stack in the middle.

Do you have any references for the performance difference between direct
hipersocket and via a TCPIP stack?

Brian Nielsen