On Thursday, 08/03/2006 at 01:17 EST, Dennis Schaffer <[EMAIL PROTECTED]> wrote: > Thanks for the sympathy. I'm beginning to feel "caught in the middle". My > network folks think the VM TCPIP stack is ancient because of how it handles > EqualCostMultiPath (and OSPF authentication, although that's now available > in v5.2) and the VM development folks think our network design is lousy > because hosts have to do routing. Sigh! As if I have any influence in the > network design ...
Hey. I never said it was "lousy". I simply said it is time to get rid of the guest LAN in favor of the VSWITCH. It's all a trade-off: zCPU and memory vs. network infrastructure expense. Some workloads will chew an entire CPU in the virtual router because of the huge volumes of data that move through it. Shop and compare. > You can still run VSWITCH in this environment. You just have to build > automation (or manually handle it) to detect the CP messages issued when > the OSA quits/resumes talking (and I'm sure there's some I haven't found in > my testing) and then configure VM and/or Linux IP stacks to IFCONFIG the > appropriate interface down/up. And, for a lot of zLinux instances, it > almost has to be automagic. The whole point of the failover mechanisms in the VSWITCH is to avoid making the guest aware of a network interruption. Two switches trunked together. Two OSAs, one to each switch. If either a switch or an OSA fails (cable, power, port), the VSWITCH automatically recovers. No fuss, no muss. > Sometimes, I wonder, considering how reliable the entire network > infrastructure seems to be, if its worth all this trouble for redundancy. > Our last network outage in the last five years was planned. Our > distributed systems have just started to utilize multiple adapters the last > two years, mainly in anticipation of that planned outage. But, just as > soon as I succumb to the temptation, it'll happen and I'll be standing > their with my so-called pants down. It happens. Repeatedly. - "What? You didn't know the switch was plugged into the electrical circuit we had to upgrade over the weekend?" - "We had to reboot the switch; I think it has a bad card." - "We promise to be more careful in the future" We just did a VSWITCH APAR last month specifically because an entire switch was taken out of service unbeknownst to the VM admin. But is it worth it? To answer that you have to have and idea of the cost of network downtime. Do your guests provide services to agents and customers? What if it dies while you're servicing customers in a natural disaster? What do you risk if those services are not available? Having the redundant configurations is only as important as the services such a configuration supports and *protects*. So only you know the answer. > I agree with your comments about routing flexibility. I haven't attempted > to move the entire guest LAN subnet but I do commonly move second-level > guests back-and-forth between either of my physical VM processors; and, the > guest LAN should work the same way as long as I move them all together. Guest LANs are fantastic. I love 'em more than anyone in the whooooole world because I no longer have to explain Subnet Theory (like String Theory in physics, but more complicated) thrice daily. They do, however, have a real cost associated with them. Alan Altmark z/VM Development IBM Endicott