craig@NNSC.NSF.NET (Craig Partridge) (10/08/88)
Hi:
I discovered at a lunch last week that there are two defintions of
dynamic IP address assignment floating around:
(1) assigning an IP address, once and for all, to a new machine
that has never been on the net before;
and
(2) assigning transient IP addresses to guest hosts (for example,
a portable PC that has been plugged into your local Ethernet for
the day by a visitor).
I get the impression lots of people are thinking about (1) and not many
about (2). Personally, I think (2) is the more interesting problem
because it makes IP much more available to travelers. (One can imagine
plugging one's PC into a jack in your airplane seat but I digress...)
CraigMills@UDEL.EDU (10/08/88)
Craig, Yummy, you stroked one of my favorite personal chords. Unpeeling another layer, the issue is whether you take your personal address with you or borrow one at your destination. The former approach is more fun, even if it does seem to require a flat address space. However, if you treat roaming hosts as relatively rare, you can map "tunnels" onto the routing fabric of your net through suitable modification of the routing tables and update message formats. Some nets, including the Phas-I NSFNET Backbone and several fuzzball nets scattered about the netscape in fact include this feature. That's how my personal fuzzball at home pretends it has addresses on three nets, including ARPANET (using transparent- host addressing), but is some hops behind the gateway to those nets. It's all done with mask-and-match tables and logical-host addressing, folks. A ubiquitous roaming feature like this could be incorporated into the Internet architecture by suitable modification of several hundred gatewasy. Only. Dave
zsu@ISTC.SRI.COM (Zaw-Sing Su) (10/09/88)
Craig, The two cases you mentioned imply different semantics for the "IP address". In case (a), the IP address is in reality a name, while it is an honest, goodness address for case (b). [re: Shoch's classic paper on naming and addressing] I second Dave that your case (b) is far more interesting. Our Reconstitution (Routing) Protocol (RP) Program supported by RADC and SAC a few years back tackled a solution to your case (b) right on. It was a pioneering effort in that direction. Prototypes were built and capability of our solution tackling your case (b) was successfully demonstrated in a series of SAC C3 Experiments. It is backward compatible across EGP to the IP Internet. Our basic approach was to separate the double semantics (as a name as well as an address) IP address carries. To facilitate address changes, we used gateway-centric addressing (in IP format) instead of network-centric as IP stands. (Gateway-centric addressing is a predecessor of Paul Tsuchiya's Landmark Routing, if you are familiar with Paul's work.) My assessment of that effort (could be very biased since I authored RP protocol) is that it was (and still is, in my opinion) a promising direction to pursue. In spite of our success, further effort is necessary to evaluate what have been achieved and how to bring it further into a practical system. Further research is clearly necessary for RP to operate in a ubiquitous mobile environment which Dave alludes to. Zaw-Sing
CERF@A.ISI.EDU (10/10/88)
Zaw-Sing, Did the RP work ever appear in RFC form? Are there readily-available papers? Vint
zsu@TSCA.ISTC.SRI.COM (Zaw-Sing Su) (10/10/88)
Vint, I don't know how I missed mentioning, in my last message, that DARPA has co-sponsored this program. Sorry! Unfortunately, we did not publish an RFC on this subject. There were two papers published in conferences, one in 7th ICCC, and the other INFOCOM 85. They were both titled "Internet Accommodation of Network Dynamics: ..." A more complete documentation of this effort is our final report to RADC/DARPA. I might still have a couple of extra copies. Zaw-Sing
morgan@Jessica.stanford.edu (RL "Bob" Morgan) (10/11/88)
Paraphrasing Craig Partridge, the two types of dynamic IP address assignment are: 1) assigning a (more or less permanent) address to a new host on the net; 2) assigning a temporary address to a host that happens to need one. Craig thinks that most people are thinking about 1) rather than 2). I think not. Far from 2) being a convenience when a visiting dignitary comes to town with their Ethernet-adapter-equipped portable PC, I think it should be the way most machines on the net get their addresses. I invite everyone who thinks dynamic IP address assignment is a simply a cute idea to accompany me the next time I have to assist a former secretary or part-time graduate student who has been put in charge of a departmental network here, and help me explain IP addresses, BOOTP tables, and default gateways. This sort of stuff is tricky yet tedious, so unskilled people screw it up, and skilled people are too busy doing something else. If we can't design these nets so that the standard end-user station can start using the net with ABSOLUTELY NO CONFIGURATION, then we have failed. What distinguishes a station as a candidate for dynamic addressing is that it supports only client-side software. I think "client-only" stations may be different enough from "hosts" (as well as far more numerous) that it's worth thinking about how their "requirements" differ (ala the ever-forthcoming RFC). Yes? No? - RL "Bob" Morgan Networking Systems Stanford
george@ditmela.oz (George michaelson) (10/12/88)
as a TCP novice I am probably way out of line here so please forgive me.
I see some problems with this area.
(1) what is the scope and validity of a temporary assignment
within the INTERnet at large?
(2) what is the threshold above which the net cannot cope with
dynamic assignment?
some peoples suggestions about what dynamic IP address assignment offer
seem (to me at least) to float functionality which requires propagation
of the new address "binding" out into the world. -If I wanted to take
a portable IP-using box around the country, and somehow get an IP address
at <new local point of attachment> and somehow propagate that back to
<my mail store host> and thus read my mail, I'd be doing things like that
wouldn't I?
-ok, one can come up with back-of-envelope schemes which support in one
persons phrase "reasonably small numbers of people doing this" but the
network-at-large has 2.5Million+ end-users. 0.1% of this is really quite
substantial traffic and address space. yes/no?
fairly simple schemes where a pool of IP addresses are reserved for new
machines already exist, Appletalk-ethernet gateways being examples.
does one connect, grab <first-free> subaddress, propagate, free, disconnect
or what? how do you adjust the life of a temporary binding to cope with
variable delay across the internet as you move around from point to point?
If I've mis-understood the intent, do please put me right!
-george
--
George Michaelson, CSIRO Division of Information Technology
ACSnet: G.Michaelson@ditmela.oz Phone: +61 3 347 8644
Postal: CSIRO, 55 Barry St, Carlton, Vic 3053 Oz Fax: +61 3 347 8987brian@ucsd.EDU (Brian Kantor) (10/13/88)
In the ham radio tcp/ip world we're facing the need for dynamic address assignment too. Since the service is radio-linked, and people often move, we have no central control over connection into our network. Nor, I suspect, do we really want such. The solution I've been toying with was to let the regional network access nodes assign the addresses. (Access by user stations/hosts to the network is to be via dedicated network access nodes, similar in concept to an IMP). When a user station fires up in an area, he requests assignment of an IP address (using some variant of RARP, I suspect), and the local access node simply assigns the next one of its allocated block of addresses to him. The address is cached and will be freed for reuse if no traffic is seen by that node from that station for some reasonable period - like six months or so. Since the underlying transport network (ham radio) has unique identifiers (callsigns), this is a relatively straightforward way to handle the problem. I can see variations on the theme: that assignment ALSO updates the distributed nameserver database; that there exists a way to flush a station's entries from the address cache; etc. With over 1500 hosts registered in over 20 countries so far, and more every day, we've got to find some way to automate the address assignment process. Brian Kantor WB6CYT UC San Diego brian@ucsd.edu