Date: Tue, 31 Jan 1995 23:55:46 -0800 (PST) From: Poul-Henning Kamp <phk@ref.tfs.com> To: hackers@FreeBSD.org Subject: SUP must die, CTM for president !! Message-ID: <199502010755.XAA19507@ref.tfs.com> In-Reply-To: <29186.791616495@time.cdrom.com> from "Jordan K. Hubbard" at Jan 31, 95 09:28:15 pm
next in thread | previous in thread | raw e-mail | index | archive | help
OK, I just had a chat with Jordan over a piece of cake (it's true !) >>> I belive that sup needs to die for all "src-current" use <<< I can still see a mission for it in the cvs-* area, where committers may want to get a up-to-the-minute copy of the cvs tree. But for src-current: It is stupid. Look at it: We run sup-scan every 8 hours, then we download individual sets of files to each user based on timestamps. That's not particular smart. We have probably all been hit by the zero byte file with the right timestamp havn't we ? Imagine the email ? "Well I sup'ed at 10:24AM (Tanzania time) from the mirror at zulu.foo.haiti and I see this bug..." Nobody can tell what version this person is running. And all the mirrors have to run supserver ? No way they will. Here is my counter suggestion: ------------------------------ We set up a machine, (thud probably) to run a shell script X times per day (X in [1..4]) which does a "make cleandist ; make obj ; make all" and if that succeeds it produces a CTM delta, which gets put in the public area on ftp.freebsd.org automatically and emailed to the ctm-src-cur mailing list at the same time. Advantages: Ftp can be used for transfer, it is much more efficient than sup I'm well connected, 6 hops to freefall, and I only use 50% of my modems bandwidth with sup.. :-( On top of that: CTM transfers only the changes, not the entire file each time a line is touched. People can ftp at the work/university and carry floppies home if they wish. People can stay current via email too. The mirrors will not even have to think about what's in the files. They can just be ftp-mirrors as usual. CTM uses md5, so it will detect if people have corrupt files. We get a sequence of "snapshots" where we know that the tree isn't entirely broken. People get a version handle they can use for reporting: the CTM-delta number. We can even make send-pr pick it up if we want to. We linearizes the load on freefall. More "customers" doesn't mean more load. Is CTM ready to do this ? ------------------------- Almost, I think. It will need a little work, but not too much, compared to the mess we would have to clean up with sup. And it should be moved to freefall/thud instead of ref. We may need a special "synchronize" version of ctm, where people can run a program, send the output and receive a special delta, which will get them up to date, but that is comparable to the load of sup today, so that should be a cheap service to run... Alternatives: ------------- Some serious improvements to sup. OK, so there, I opened my big mouth again. I hope to get support from some of the people who ran or runs CTM... (What needs to be better ? You guys are remarkably silent ! speak up ! Now is the time to retrospectively get on the cutting edge :-) Will anybody be willing to step in and help make CTM perfect ? I will clearly not have all the time to do this myself and be the 2.1 release engineer at the same time... I wouldn't mind handing the thing over to somebody else entirely if somebody were to step forward... (HINT HINT !) What do you think ? -- Poul-Henning Kamp <phk@login.dknet.dk> TRW Financial Systems, Inc. FreeBSD has, until now, not one single time had an undetected error. :-)
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199502010755.XAA19507>