On 2011-11-28 13:34, hvlems at zonnet.nl wrote: We're all in the same boat, using a protocol, hardware and o/s'es from days long gome :-) so I assumed that shariing would be beneficial and thus desirable. True. But it all boils down to how much effort it is worth. The practical side is obviously a lot more complex. Yes. Every site starts with area 1. Merging them is not trivial. In our case even a simple solution like "one group adds 512 to its nodenumbers" won't work given the ranges in use. Right. And once you start renumbering, it makes little difference if we renumber within one area, or change to another area. It's about the same work. But my concern here is that it's way complicated to get everyone in area 1 on either side to renumber at a somewhat same timepoint. And without that, we face larger disruptions for a longer time. Renumbering one area is one thing. Modifying DECnet exec, line and/or circuit parameters is something else again. Hmm. Right. Since the italians are tweaking all kind of parameters, it might be a real issue to get everyone on the HECnet side to tweak them in a consistent and working way. And a bridge is really not the way to go to get a really huge network. Ethernet segments that become really big will cause a *lot* of broadcast traffic going everywhere. It grows with the square of the number of connections, more or less. The more I think of it, the more I suspect that hooking the italian DECnet straight to HECnet through a bridge is not a good idea, from the italian side. We really need a router in between, to not swamp them with traffic. For the record: I operate 40 real VMS systems, two virtual ones and five or six pc systems that run decnet. That's a lot of work to maintain. Indeed. You are as bad as Saku... ;-) What worries me most is the performance issue. The meshed concept is a good idea, so it seems, but might possibly be in need of an improved implementation. UDP is connectionless and hence less robust than TCP but it shouldn't drop data. So that needs attention IMO. My main worry would not be performance, unless we go down TCP, in which case it will be a real problem. For UDP, if something along the way can't handle the traffic, UDP packets will be dropped, which is good. DECnet have mechanisms to recover lost packets, and will slow down and retry. TCP does all of that itself, which means that if a downstream node can't handle the traffic, you will get TCP congestion, which will block TCP writes, or else give write errors at the source. And you'll have a whole backlog of data in the queue that will eventually be delivered, but it will totally mess up the DECnet timers and retransmits. A meshed solution will not really improve performance, unless you also implement dynamic rerouting, so that it can send packets the shortest path to destinations. Otherwise it is just more of a stability enhancement, since if one link is down, there is another to use. However, this is the interesting question: how does the implementation done figure out when to switch to an alternative link? You do not want to have two links active at the same time, since then you create a loop, which is the *very bad thing* I don't want. Johnny Hans -----Original Message----- From: Johnny Billquist<bqt at softjar.se> Sender: owner-hecnet at Update.UU.SE Date: Mon, 28 Nov 2011 13:05:56 To:<hecnet at Update.UU.SE> Reply-To: hecnet at Update.UU.SESubject: Re: [HECnet] Integrating with the Italian network. Actually, connecting the italian guys shouldn't be that complicated. If they just have one point with a fixed (or atleast somewhat static) IP address, we can hook up at that point. Neither side should need any other modifications. The renumbering is another issue. Unfortunately, while still maybe simpler on our side, area 1 is still the one area with the most number of different stakeholders on this side as well. It is a bunch of people, apart from me... I've done something similar, giving out subranges to people in area 1. I suspect a merge might actually be easiest. Move some machines around on individual basis and keep some machines around. As for changing the address in RSX, yes, once the config have been changed, I need to restart DECnet. I might need to reboot as well. Either way, that is simple. It's just lots of work that needs to be done before we can do a merge. But first we need to decide if this is something we want, before we start work. By the way, it might actually be smarter to not bridge us together, but instead let two VAXen run a connection instead, since that will avoid the bridged ethernet from becoming silly large. Since Gerry indicated that traffic volumes are a concern, as well as packet size, the bridge is really not a good solution for them. If they can split their network into several sub-segments, they can drastically reduce network traffic, and by having a VMS machine with DECnet over IP used as a router in between, they will not increase the traffic at all, for most sites, if they were to hook up to HECnet. I should check how they did the "full mesh" thing. Since this is one annoying problem with the bridge, if they have a good solution to it, maybe it would be something to implement. The normal "correct" solution would be to implement the spanning tree protocol in the bridge, but I've felt too lazy to do that before... Johnny On 2011-11-28 12.08, hvlems at zonnet.nl wrote: Gerry, Thanks for the clear explanation of the retro DECnet setup. I agree with your evalluation of renumbering area 1, yours seems to have more system owners than HECnet's area 1. Yes, in terms of planning it was a lot easier to modify just my own systems. The complexity of such a project increases sharply when the number of stakeholders goes up :-) About the bridge program, I mis understood the DNS part: hence the reference to phase V. The IP part was understood and I still think it tries to solve issues that ought to be done elsewhere. I am a strong believer in the KISS principle. Which is why I favor Johnny's original version. Perhaps we can still merge the two, provided you guys want this and Johnny wants to move out of area 1. Its not complicated just a lot of reboots. I did very little with DECnet on RSX but I think a reboot is all that is needed. Hans -----Original Message----- From: gerry77 at mail.com Sender: owner-hecnet at Update.UU.SE Date: Mon, 28 Nov 2011 02:59:45 To:<hecnet at Update.UU.SE> Reply-To: hecnet at Update.UU.SESubject: Re: [HECnet] Integrating with the Italian network. On Sat, 26 Nov 2011 12:05:30 +0100, you wrote: networks. "Ownership" of area 1 may possibly have ego involved though I doubt that Johnny would care much. That's not our case. We just picked area 1 because it was the most "natural" choice. At first many of us didn't bother about DECnet at all and choose the first available value, i.e. 1.1. Then, when we started dreaming and thinking about a network, we started choosing address ranges in the 1.1 to 1.1023 range: one location from 1.1 to 1.9, another from 1.10 to 1.19, and so on. Each location (not always a physical one) was assigned to an administrator, and when that administrator filled its ten-addresses range, it received an additional one, not always contiguous to the previous. When the network became reality we implicitly choose to keep that numbering standard. We have also a few exceptions: the first was our Simh VAX/VMS V4.7 node that earned 1.666 because was evil to set up to our needs; the other is 1.1010 for the TOPS-10 node which also got DIECI as its node name (because dieci is the italian word for ten: compare it to decem in latin). There was a time when we considered changing our area number to 39 (the international dial code for Italy), but then there was really no need to do that and some members of our network are somewhat difficult to reach and speak with, so we would have ended with a mixture of areas 1 and 39, and that led to the cancellation of the renumbering project. When you, Hans, renumbered your systems, you were just one single person renumbering his personal nodes, under your direct control. We are in a situation which is more like asking the whole HECnet (or a major part of it) to change numbers. In another message on this thread I have posted a link to our node list. In that list there is a column showing the administrator nickname for every node. Well, that column does not tell the real story: there is a good number of nodes administered by one person, but physically located remotely from that person. Some nodes are active, but some others are switched off and do come online from time to time, without prior advice and it's not so easy to tell the relevant people to remember not to come online without prior intervention from some of us in order to change the DECnet address. So, go figure what would happen if someone comes online with e.g. the MIM address. What I've read about the "enhancements" to the bridge program is yet another matter. First off, it ain't called bridge for nothing: like any layer 2 bridge the program moves packets between ports and is transparent to both the content of the datagrams and the functionality of the protocol. Including DNS like functionality violates that rule. In my opinion, if anyone wants a central name repository for DECnet please upgrade to DECnet phase V. Which is my way of saying that I'm not too fond of this "enhancement"... I think you have misunderstood our DNS functionality. Our bridge just tries to resolve bridge endpoint IP addresses, but it is completely DECnet transparent, i.e. it does not and never did try to resolve DECnet node names into DECnet addresses. That's a service specific to DECnet Phase V, as you correctly appear to suggest. If our bridge does not receive anything (that is neither data nor even a hello packet) from an endpoint defined as dynamic, after a certain amount of time it tries to resolve again the hostname of that endpoint only. Because our bridges are transmitting and receiving most of the time, the resolve routine gets called quite seldom, so the delay issues explained by Johnny may be encountered as seldom as an IP address change, and we consider it a honest "price" to pay to have automatic dynamic IP addresses working. Bye, G. P.S. please, forgive any syntax or grammar error: I'm not English native. . -- Johnny Billquist || "I'm on a bus || on a psychedelic trip email: bqt at softjar.se || Reading murder books pdp is alive! || tryin' to stay hip" - B. Idol