Date: Mon, 20 May 1996 22:05:40 -0700 (MST) From: Terry Lambert <terry@lambert.org> To: msmith@atrad.adelaide.edu.au (Michael Smith) Cc: jimd@mistery.mcafee.com, gpalmer@FreeBSD.org, msmith@atrad.adelaide.edu.au, slagos@net1plus.com, questions@FreeBSD.org Subject: Re: <no subject> Message-ID: <199605210505.WAA29942@phaeton.artisoft.com> In-Reply-To: <199605210243.MAA24367@genesis.atrad.adelaide.edu.au> from "Michael Smith" at May 21, 96 12:13:55 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> > I currently have two FreeBSD boxes mirrored as ftp sites. > > I'm using DNS round-robin to balance the load. The question > > has come up: "What happens if one of them goes down?" > > The connect to the first address should fail, and the client should > move on to the next. This, unfortunately, depends on clients being > implemented correctly (which is unlikely). Yep; it means you are probably screwed. And we haven't even talked about caching keeping the old address as the only one you are allowed to resolve to. 8-). > > The idea is that machines in a round robin ring could > > "cover" for one another using simple scripts and IP aliasing. > > There's no kernels hacks, no dynamic DNS hacks, and no > > applications layer hacks necessary. > > Cue terry to drop in and talk about server vs. service connections. 8) ...Mike means that it wants a protocol change at a pretty fundamental level (the "what is a connection?" level). Even IPv6 is screwed here. Wait for IPv7 (~8 years unless someone does something right); it should be pretty obvious to everyone in about 2-3 years... in some places on the net, it's becoming obvious now (cv: the MAE-West NASA FDDI link that was choking to death ~5 weeks ago, and the currently overloaded MCI backbone, choking to death daily at a node near you). I'm thinking on something I call "MNTR" (prononced 'Mentor', in honor of Billy Batson 8-)) for Minimal Network Transit Routing. It relies on data vaults for regional service replication to offload backbone traffic. In my ideal world, you pay for your pipe flat rate based on pipe size, and people putting up resources pay vaulting services for regional replication -- leaving room for volunteer vaulting and self-vaulting (with source quench) for small sites or distribute free sites. Long hop routes should be reserved for important things which *must* have central servers badly enough that they merit taking up overall network resources. For instance, netrek, but not MUDS or IRC. 8-) 8-) 8-O. Just joking; actually, intra-service cluster links will probably need to have backbone channels as well., and IRC and MUDS qualifiy better than netrek (not being nearly as interactive as 3-5 updates a second). IPhone is another candidate. If you are really interested, you need to consider service referral records (like MX records) and participate in the recently formed FTP protocol revision mailing list. This should stave off the wolves for another 2-4 weeks 8-). Note that they haven't thought of this on their own -- they're mostly interested in making the list command deterministic for GUI-based tools, so they can put up folders and crap on a normal desktop. You'd have to elect yourself Mr. Advocate. In the long run, it probably won't matter what you do to ftp itself, as the net becomes swamped in http, video, and IPhone traffic. Regards, Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199605210505.WAA29942>