[68] in Zephyr Mailing List
Fixing the Zephyr daemon
daemon@ATHENA.MIT.EDU (Doug Kingston)
Tue Jun 2 19:48:06 1992
Date: Tue, 2 Jun 1992 19:44:18 -0400
From: dpk@fid.Morgan.COM (Doug Kingston)
To: zephyr@Athena.MIT.EDU
Cc: technology@gateway.morgan.com
I have been talking to a number of people who have been looking
at Zephyr and we have all reached a similar conclusion. Namely that
the functionality is almost exactly what we need at our sites, but
the zephyrd braindump protocol prevents it from being useable
with more than 2 daemons. I would like to work with others to shake
down this problem once and for all.
I have in interest in running perhaps 15 daemons. This number is arrived
at because we have 7 different "clusters" scattered around the world in
Paris, London, New York (2), Chicago, San Francisco, and Tokyo. We need
the zephyr daemons to local to the clusters of workstation to keep
unnecessary communication off the long-haul links, and a second daemon
at each location to cover failure of the inital daemon.
The fundamental problem with the current daemon is that it isn't multi-
threaded, inplicitly or explicitly. So, when it goes to brain dump with
some one it "goes off line". This is also true when trying to connect
to another node either to send or get a braindump. If the other party
isn't there, the daemon goes catatonic waiting for the connect to
time out.
With our current operating systems, we have all the hooks to do this
right and keep full connectivity in place. Since we will likely be
making major surgery to zephyrd, I would like to throw the table
open to others comments. In particular, who may be already working
on this or have ideas as to how it might be re-architected.
If this looks like something that isn't directly on anyone's plate
I will be happy to help coordinate the effort of reworking the daemon
either from scratch or in a remedial fashion. I am currently leaning
towards a complete rewrite.
Comments?
-Doug-
PS. I have no intention to introduce protocol changes. I am strictly
interested in an implentation of the zephyrd function that
scales up to quite a few servers.
