Lauren Weinstein <lauren@vortex.UUCP> (08/04/83)
The use of "sender" sitenames in uucp filenames definitely should reduce the chances of filename collision. It should be noted, however, that on the sender systems uucp's are free to use whatever filename formats they desire, since it's only the filename appearance on the *remote* system that really needs to be standardized. For systems that run with separate subdirectories for each site, using the sender's name for all filenames on the sender system is probably OK, since the subdirectory itself uniquely specifies the destination for the files. On systems that run only one /usr/spool/uucp directory, or single C., D., X. subdirectories, it might make more sense to store the files locally in "remote name" form but send them to "sender name" files on the remote side. The whole point here is to make it easier for a system person on the sender system to easily tell which files are destined for which sites when manually inspecting the queues. I suspect that this wouldn't be such a bad idea for sites running with separate site subdirectories as well. The vortex UUCP, for example, runs with separate C., D., X. directories. At the next compile, all vortex C. files will be of the following form: (only some fields are shown -- sample is for mail to "floyd") S D.floydAQRST D.vortexAQRST S D.floydXQRSR X.vortexXQRSR As you can see, this arrangement leaves the files sitting locally with the remote system's name, but delivers them with the local (sender) system's name to avoid filename collisions and allow proper X. file sorting. --Lauren--
henry@utzoo.UUCP (Henry Spencer) (08/05/83)
There is one serious problem with Lauren's D-file naming scheme, and
I dearly hope Lauren is not going to imbed it into any production
version of uucp. The problem is, it makes the collision problem WORSE.
(Since this was the biggest objection to the old naming scheme, this
seems a backward step.) With Lauren's scheme, outgoing files and
incoming files on a given site have (potentially) the same names!
A file named "D.vortex..." on vortex may be either an outgoing file
waiting for transmission or an incoming file waiting for disposal,
and only the quasirandom sequence numbering prevents collisions.
This may not be too serious a problem for a leaf node, but for a
site that does heavy relaying it's a disaster. Please DON'T DO IT!
More generally, some of the naming problems could be solved by working
with the character position now used only for grade letters, but only
if it is a network-wide standard. Could we PLEASE have an end to
surprise unilateral changes to things that affect compatibility across
the whole network?!? There have already been some very unfortunate
precedents set. If we keep on this way, we'll soon be split into a
set of incompatible subnets that break each other's uucps when they
try to talk to each other. Please, people, show some consideration:
think for a moment about compatibility and about the problems other
people will have before you implement your latest bright idea, no matter
how wonderful it is.
--
Henry Spencer @ U of Toronto Zoology
{allegra,ihnp4,linus,decvax}!utzoo!henry