home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cis.ohio-state.edu!pacific.mps.ohio-state.edu!linac!att!ucbvax!ATHENA.MIT.EDU!wdc
- From: wdc@ATHENA.MIT.EDU (Bill Cattey)
- Newsgroups: comp.soft-sys.andrew
- Subject: Re: "slow networks"
- Message-ID: <IebJxxoGf0478PRKEF@athena.mit.edu>
- Date: 27 Aug 92 23:04:29 GMT
- References: <QebI70G0ts4j8lz1Yy@alw.nih.gov>
- Sender: daemon@ucbvax.BERKELEY.EDU
- Distribution: world
- Organization: The Internet
- Lines: 29
-
- It turns out that there is a great deal of hair in the message server,
- and a great deal of AFS version 3 specific code there. The abstraction
- barriers are regularly crossed in complicated but necessary ways to deal
- not so much with security as reliability.
-
- The AMDS mail delivery method is fundamentally flawed. It is bad design
- to require the delivery agent to have to deal with the possibility that
- there are as many delivery agents as there are users all attempting to
- write into a unix file system home directory to copy mail in and update
- status simultaneously.
-
- Client/Server is a good model for a mail delivery agent. It puts the
- smarts for dealing with race conditions into small well-understood
- blocks of code. It does not require a high speed network connection at
- all times for mail delivery.
-
- The AMDS method was an interesting experiment, and a lot of people
- worked their butts off making it work and making it reliable. I expect
- they know more about race conditions than any other mail system delivery
- gurus. But the system would benefit greatly from a paradigm shift to
- the common client-server models.
-
- The toolkit architecture is another experiment. But this experiment
- turns out to be well suited to the problem domain and will continue to
- influence GUI and toolkit designs for several years. I believe that the
- ATK richness should be kept, but that the AMDS richness should be
- learned from and then radically simplified.
-
- -wdc
-