ken@dm.unibo.it (05/08/90)
I have run into additional minor problems with V2.0 in Bologna:
1) When building metav1.2b on the sun3, you need to fix the symbolic
link metav1.2b/sun3/lib/sal/Makefile. It currently points to
outer space and should be changed to point to ../../lib/sal/Makefile
2) Also in meta, many references are made to
../../../isisv2.0/lib/lib1.a lib2.a and mlib.a
this mix a new and old name incorrectly. The new library names are
.../lib/libisis1.a libisis2.a libisism.a
and the old lib1.a, lib2.a and mlib.a exist only in clib.
Making links from ...lib/libisis1.a to lib/lib1.a and so forth is
the easiest fix. With this change and (1), both meta and magic
compile easily.
3) When I built ISIS with bypass enabled and ran grid in the random mode
or in gbcast mode, I managed to get it to hang after printing a
message about an "unexpected flush message" arriving in the wrong
view. This happens only when several gbcast's are done at the same
time in a group that is actively receiving bypass cbcasts, so I doubt
that it will affect people. I will post a bug fix when I have one.
The same (or a related) bug can be triggered if you kill off a
process and a mix of cbcasts and gbcasts is active at the exact same
time.
grid with just cbcast or just abcast runs incredibly fast, but I did
notice that with one member in cbcast mode, grid is _very_ slow to
notice group joins and only a tiny bit quicker to notice when you
select the stop button. This application, with 1 member, is
basically an infinite loop of X11 updates and apparently doesn't
check for isis input events often enough. An example of a trivial
fix is to call isis_accept_events(0) inside the loop (in grid, this
would be in the routine called "receive(mp)"). We'll work on this,
but it isn't really an ISIS bug... I would call it more of a bug in
the grid demo itself.
4) For some reason, the .ps version of the manual was never put into the
distribution areas and the .dvi file may be hard to print (it assumes
the Cornell defaults for font and point size).
... all in all, I think we are off to a reasonable start on V2.0. I'll
fix bugs and release some sort of patch list when I get back to Ithaca.
Ken
[ By the way, I am spending two weeks visiting in Bologna. What a
[ beautiful city! If anyone is hesitating about Bologna `90, this
[ year's version of the Advanced Distributed Systems Course in which
[ I taught last year and the year before (Arctic 88, Fingerlakes 89),
[ let me strongly recommend that you enroll. The lecturers are super
[ and what a great place to spend ten days! This year's organizer is
[ Prof. Ozalp Babaoglu: ozalp@dm.unibo.it, and the registration deadline
[ is approaching. The course is a very comprehensive introduction to
[ the whole field of distributed systems and was a real success in 88
[ and 89. July 23 - August 2nd in Bologna.
far...