On Wednesday, 10/25/2006 at 12:45 ZE2, Chris Mason <[EMAIL PROTECTED]> wrote:
> The E005 case is obvious if any thought is given to passing IP addresses but > the E00A case is not so obvious so thanks for the elaboration. No problem. I just wish Device Drivers Everywhere would stop displaying the [EMAIL PROTECTED] reject code and actually generate a meaningful message.... > So why do you say "in theory"? If I have described it correctly - also I > hope an improvement on the way it is described in "System z9 and zSeries > Open Systems Adapter-Express Customer's Guide and Reference"[1] - then it's > very clear what should happen and surely what does happen. "In theory" == "Bugs happen". > The question that now arises is how to reconcile the "gratuitous" ARP which, > in essence, boldly asserts ownership of an IP address with an ARP request > which, in essence, timidly inquires whether or not an IP address is already > in use. Unfortunately I don't have testing facilities available to me these > days in order to sort out these sorts of questions. > > Probably I've overlooked something obvious which testing would immediately > show me. IIRC, when the OSA is instructed to "takeover" an IP address, it applies only to the check for other MPC groups on the same chpid (E005). > Another topic: you are implying - I'm sure you didn't mean to - that there > is but the one HOME address. There are of course potentially many HOME > addresses potentially needing to be checked when received by an OSA. For > example, the IP address could be associated with a dynamic VIPA that "moves > around" between IP instances. We see such a VIPA in Pierre-Andre's case. I > never had the wit to try having VIPAs belong within the address range, > subnet, associated with the immediately adjacent LAN but it seems to be > considered a valid "trick". I was trying to keep it simple. As you can imagine, when you have two adapters connected to the same LAN segment, it isn't quite so straightforward to manage the VIPA adddress. Once registered in the first adapter, the second adapter will discover it is already assigned (E00A). But, wait!, that's no good. So the registration is done WITHOUT the ARP and the grat ARP. It is simply added to the list of registered addresses. > By "host" in your last paragraph you are specifically referring to an > interface address on the LAN on which the relevant OSA port is also an > interface. Sorry for the pedantry; it's a habit from teaching students who > do not have English as their mother tongue. Indeed. If there are duplicate IP addresses active, but they are for some reason not in the same LAN segment, then more serious problems are afoot. So, I discard that case from discussion and assume that the subnets are correctly managed, even if a host is not. Alan Altmark z/VM Development IBM Endicott ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html