On Wednesday, 05/20/2009 at 03:55 EDT, Tom Duerbusch <[email protected]> wrote: > When mainframes just started getting into IP communications, one of the things > we were warned about is to filter out any unnecessary IP traffic that is not > for the same subnet that the mainframe was on. The reason that I remember, is > that the IP stack looked at every packet to see if that packet was destined for > itself. Hence an interrupt was generated for each packet sent on the LAN. > > Ok, now we are on VSWITCH and Guest Lans. Does that same problem exist within > the VSWITCH/Guest LANS?
For Layer 3, VSWITCH/Guest LAN, no. For Layer 2, it depends on how bright the switch is. Modern switch gear learns what MAC addresses are associated with what ports and will, over time, adjust traffic so that ports only receive traffic they can use. Once the traffic in CP, he sends it only to the guest to whom the packet is intended. > I don't have a performance monitor, so I don't see a problem <G>, but I wonder > if this could become a problem. > > Right now, I have plenty of CPU. I can afford to waste it. In the future, > that might not be the case. I'm wonder if I'm going down a scary path, that is > just going to get darker and darker, until.... Since you don't have a performance monitor, you may not know if it becomes a problem until too late! Your network switch performance monitors should be able to report on percentages of unicast/multicast vs. broadcast traffic. It is only a broadcast that will wake up everyone unconditionally. (That's one of the benefits of moving to IPv6 - no more link-level broadcasts!) Alan Altmark z/VM Development IBM Endicott
