On Mon, Apr 5, 2010 at 10:47 PM, Alan Altmark <alan_altm...@us.ibm.com> wrote: > On Monday, 04/05/2010 at 04:35 EDT, "McKown, John" > <john.mck...@healthmarkets.com> wrote: > >> So you'd need a "router" machine to talk between the hipersocket to z/OS > and >> the VSWITCH, right? > > Yes, which undoes all the performance benefit of HiperSockets by funneling > all the traffic through a single guest.
And don't forget that talking to a virtual NIC is more expensive than to a real subchannel. > z/OS needs only 3 addresses per HiperSocket chpid, so I don't understand > why adding more guests to existing chpids is creating a z/OS problem. Right. And giving z/OS access to an OSA chpid takes approximately the same number of UCB's... There's several other situations where you should not use hipersockets, but this can't be an issue. Looks to me someone is trying to find a problem for his solution... Rob
