ken@gvax.cs.cornell.edu (Ken Birman) (07/13/90)
We seem to be getting close to a V2.1 "bug fix" release. Here's
a snapshot of what we have been doing:
1) SIGALRM related problems have been fixed by eliminating use of this
signal in V2.0
2) Fortran problems have been fixed; mostly due to arg passing by ref.
3) A memory leak associated with pg_lookup has been fixed
4) BYPASS mode doesn't freeze up anymore, and hence will now be useful.
5) A memory leak in the meta table manager has been corrected
6) Several problems in the long-haul and spooler facilities have been fixed
7) We fixed some problems that relate to scanning sites files in isis.c
8) Some of the include files had problems with function declarations
9) "NACK" messages were failing to trigger prompt retransmission of lost
packets.
10) potential filename "overflows" in some utilities have been eliminated
11) rexec logs were not written into the logdir.
12) fixed the magic lantern code (not really part of ISIS)
Things we are still working on and expect to fix momentarily
1) "news" sometimes jams up
2) isis -Z sometimes doesn't shut the system down
3) possible problems with isis_remote between the MIPS and HPUX systems
4) slow group create on HPUX systems with large numbers of sites
(the initial create seems to take unexpectedly long sometimes)
5) under some very peculiar conditions, rexec can hang (hard to reproduce)
6) rmgr sometimes starts up "too quickly" and gets confused about the
incarnation number of remote sites, causing it to hang.
7) the system fails to notice if you mix bypass and non-bypass processes
in a single group (it hangs without complaining, should "panic")
8) process lists have not been tested adequately.
9) isis_transport has not been tested adequately.
When we get to the bottom of this list we'll do the V2.1 "release".
Note that V2.1 won't have any new code, just fixes to old code.
We are hoping, though, that starting with V2.1 it will be possible
to enable BYPASS by default. This will eliminate many of the congestion
problems the current release exhibits and make life much easier for those
of you designing performance intensive applications.
Things that will be done by the end of the summer or during the fall:
1) pg_client mechanism that runs via the BYPASS code (fast protocols for
client-server and "diffusion" groups).
2) the hierarchical group tools
3) the Meta Lomita language
4) Deceit
5) perhaps an ethernet multicast driver hooked into isis_transport
6) perhaps an ISIS/VMS link. (ISIS-VM/CMS is on hold right now)
It is unclear just how the academic and commercial sides of ISIS will
interact, so I won't make promises about releasing this code in source
form from Cornell right now. However, anything that works well enough
will eventually be released through one entity or the other (depending
on who developed it) and whenever possible, we plan to release some useable,
free version of any system developed at Cornell.
Longer term plans: don't hold your breath, but we're heading there
1) a "kernalized" version of the toolkit
2) (1) in a very high performance setting, like Mach or Chorus or both
3) a realtime toolkit
4) more complete long-haul tools