alext@ccwf.cc.utexas.edu (Alex Tomlinson) (02/28/91)
Does anyone have a list of known bugs in V2.1, or know where I can find one? thanks, Alex Tomlinson
ken@gvax.cs.cornell.edu (Ken Birman) (02/28/91)
In article <ALEXT.91Feb27191304@doc.cc.utexas.edu> alext@ccwf.cc.utexas.edu (Alex Tomlinson) writes: > >Does anyone have a list of known bugs in V2.1, or know where I can >find one? > We never bothered to keep an accurate list, but here is a rough list of things that have been fixed in V3.0: - cl_dump gave bus error when application combined isis_input with isis_select - chmod(0x666,... ) in pr_client.c prevented connection to ISIS in unix-domain - overflow of ISIS_TIME caused protos to enter infinite loop after 24 days - gbcast messages lost in bypass mode - view incarnation numbers "skipped" in bypass mode - bcast_l 'x' didn't work in bypass mode - forward() broken - no error codes returned on asynchronous multicasts - bcast_l 'x' returned wrong value of isis_nsent in non-bypass mode - bcast_l 'x' dumped core in bypass mode when bcast was asynchronous - isis_remote sometimes sent null-replies on behalf of live, non-remote processes (confusing the sender, for obvious reasons) - several performance problems in bypass mode, mostly due to transmission of extra ack packets or premature transmission of "termination" packets, both of which slowed things down (but were not incorrect, per-se) - "meta" jammed up in bypass mode - tab not permitted in "sites" file, treated as a text character - isis_connect broken for certain connection sequences (hangs) - memory leak in meta table manager - long-haul transmission code didn't recover when line broke during file transfers (worked OK for other cases, though) - spread sheet X11 interface didn't work on color displays - bcast_l 'f' option hangs if more than one broadcast is forked off at a time - token tool didn't correctly recover from failure of the token holding process -- for some failure sequences, token vanished - rexec utility capable of starting up "too quickly" and hanging; subsequent rexec requests would start the desired process but the call to rexec() itself wouldn't come back (infrequent, timing dependent) - in bypass mode, sometimes had a 1-second delay when application that was sending in group A switched to group B Mark Wood probably has a separate list of bugs he has fixed in Meta ... this may seem like a long list, but it is probably about par for a system of this size. ISIS is very large and most of these bugs are pretty obscure -- and more than half only arise if you enable BYPASS mode. The others involve ISIS options that most users don't work with. When I say that these are fixed in V3.0, I mean that the V3.0 release (available now) doesn't have any of these problems. Of course, it will probably have other bugs, but ISIS Distributed Systems Inc. will be providing professional support and keeping a careful bug log for its users. V3.0 is also being tested more extensively than any previous ISIS release. Here at Cornell, we just haven't had the resources to do that sort of thing while also working on our new system and writing papers. As readers of this group are aware, we do plan a V2.2 release, and we will fix most of these problems. However, since V2.2 will be our last public release, we plan to wait until we are seeing no additional bugs at all before generating it. And, we are pretty busy now. So, I see V2.2 coming out at the end of summer of 1991, at the earliest, and possibly later than that. And, of course, it will only fix bugs: any genuinely new features added in V3.0 won't be included in V2.2, because most of the new code is proprietary. Please don't misunderstand: we would like to provide free support, but the number of ISIS users has grown substantially and providing support is getting too expensive for our research budget to underwrite. We are very pleased to see people using ISIS, and through ISIS Distributed Systems, we plan to see the system continue to develop and become increasingly robust -- most users are finding it remarkably solid, or so they tell me. However, it just isn't possible to get public funding to do a professional job of maintaining and supporting a system like this. If you haven't received information about the commercial release and do need support, send your postal address to one of us and we will have the company send something out. -- Ken Birman ken@cs.cornell.edu