home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!zephyr.ens.tek.com!uw-beaver!cornell!ken
- From: ken@cs.cornell.edu (Ken Birman)
- Newsgroups: comp.sys.isis
- Subject: Re: Process group addresses, DCE etc.
- Message-ID: <1992Sep2.162739.23713@cs.cornell.edu>
- Date: 2 Sep 92 16:27:39 GMT
- References: <barry.715250287@citr.uq.oz.au> <1992Aug31.185652.21586@cs.cornell.edu> <barry.715390460@citr.uq.oz.au>
- Organization: Cornell Univ. CS Dept, Ithaca NY 14853
- Lines: 75
-
- In article <barry.715390460@citr.uq.oz.au> barry@citr.uq.oz.au (Barry Kitson) writes:
- >Hi,
- >
- > Do all clients of a process group have the same address
- >for the group? What I don't understand yet, is how virtual synchrony
- >is implemented, and what protocols operate between which objects.
- >How is ordering controlled in bcasts and who is responsible for
- >controlling it?
-
- Isis implements protocols and a model described in two ACM TOCS papers,
- one in August 1991 and one in June 1987. A prior ACM SIGOPS paper, also
- in 1987, describes the toolkit and the overall approach -- although TR1216
- will eventually be published and does a better job than the original 1987
- paper, I think.
-
- To answer your questions, though: yes, all processes in the system that
- do a pg_lookup on a group (within the same LAN) get the same internal
- ISIS "address" back. ISIS knows how to expand an address into the
- current membership of the group, but this information is only stored
- and updated at the members of the group, so anyone who isn't a member
- gets data that could be stale by the time it arrives. We have a whole
- bunch of protocols engineered for different communication patterns and
- ordering properties, and you should read the papers if this interests you.
- Note, however, that the new ISIS system will actually be implemented
- entirely over cbcast and a type of "flushed" cbcast that achieves
- the properties of gbcast. This is discussed in a TR by me, Cooper and
- Gleeson (isis publications list appended). So, we don't really need
- a complex implementation, it turns out, to support a very flexible
- protocol suite.
-
- Isis is totally decentralized, although some protocols do pass tokens
- around that allow the holder to do some distinguished thing, like
- ordering two concurrent events. But, when we do this, the token is
- automatically regenerated in the event of system reconfigurations.
-
- I think that the more I say on this the more confusing it will get,
- short of pointing you again to the detailed model and discussion in
- the journal articles, which were written to be precise.
-
- >>No, we leave this to the OSF and UI Atlas people... Sounds like you need
- >>DCE; then you can ask "can ISIS be used within DCE" and we can reply "sure".
- >>(This is no problem...)
- >
- > How do Isis and DCE fit together? Do you make use of DCE RPC's for
- >secure communications? Does Isis use DCE threads? (I don't know much about
- >how DCE offers all these services to applications programmers, so don't assume
- >too much.)
-
- The fit is pretty much one of "non-inteference". ISIS programs in
- a DCE environment would use the POSIX threads package, aka pthreads,
- which ISIS already supports when compiled for MACH. (Actually, we might
- need to edit some include files a little once we get ahold of a real
- DCE distribution, but this will not be a problem). The isis routines
- would then be running concurrently with your "normal" DCE application.
-
- RPC would be into your application, through DCE RPC, but the various
- ISIS primitives could all be called -- so you basically just get both
- worlds at the same time. Since DCE has encyption facilities, you could
- invoke these primitives to encrypt data sent in ISIS messages...
-
- ISIS itself doesn't use DCE directly, except for the threads package.
- In fact, we can use any of a half-dozen threads packages, but our model
- is pretty much the pthreads one, so DCE is a particularly close match.
-
- By the way, we are working closely with OSF's Research Institute on
- adding ISIS functionality to OSF 1/AD. When this is done, we plan to
- have a look at how that functionality might be used within systems
- like DCE. (We aren't particularly biased towards OSF -- we are
- also working with everyone else!)
-
- Ken
- --
- Kenneth P. Birman E-mail: ken@cs.cornell.edu
- 4105 Upson Hall, Dept. of Computer Science TEL: 607 255-9199 (office)
- Cornell University Ithaca, NY 14853 (USA) FAX: 607 255-4428
-