home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.isis
- Path: sparky!uunet!usc!rpi!batcomputer!cornell!glade
- From: glade@cs.cornell.edu (Bradford Glade)
- Subject: Lightweight Process Groups
- Message-ID: <1992Sep7.170427.22271@cs.cornell.edu>
- Organization: Cornell Univ. CS Dept, Ithaca NY 14853
- Date: Mon, 7 Sep 1992 17:04:27 GMT
- Lines: 118
-
- >Newsgroups: comp.sys.isis
- >From: ken@cs.cornell.edu (Ken Birman)
- >Subject: Re: ANSA, ISA and ISIS.
- >
- >In article <barry.715406765@citr.uq.oz.au> barry@citr.uq.oz.au (Barry Kitson) writes:
- >>Just found another interesting snippit in the Isis Ref. Man.
- >>
- >> What is the current status of ISIS wrt ISA and the
- >>ANSA architecture?
- >>
- >
- >I really like the ISA/ANSA work, but our ties with that organization
- >are less active lately. I tend to see ToolTalk and NewWave as the
- >commercial realizations of ANSA (aka OMG) and we have been thinking
- >about how to integrate ISIS better into this type of OO environment.
- >
- >We have a paper, with Olof Hagsand (olof@sics.se) and Holger Herzog
- >(Siemens) on issues raised by this problem. The result of this study
- >was a prototype OO presentation of ISIS for C++ users, and also a lot
- >of insight into the role that lightweight groups will play in future
- >versions of ISIS. Brad Glade is now working on this problem at the
- >level of operating systems interfaces; the plan is to then work up to
- >higher level interfaces and finally focus on integration with the OO
- >environments sometime next year.
- >
- >Brad -- could you say something about lightweight groups? People probably
- >won't know what this is about...
- >
-
- Sure ...
-
- Lightweight groups in Horus (aka ISIS++) are designed to address the
- performance problems encountered by ISIS applications programmers that
- use lots of highly overlapping groups. Casual use of the process
- group paradigm in ISIS applications can incur unnecessary overhead
- when maintaining causality and failure atomicity obligations across
- groups. When a failure of a process occurs, ISIS must flush active
- (unstable) messages for each of the groups to which that process
- belongs, and deliver a new view of the group's membership.
-
- The impact of this protocol when taken alone is fairly small
- (essentially forcing synchrony at all members), but when combined with
- other concurrent invocations of the protocol, it can have a large
- impact on the system, due to the large number of messages being sent.
- Unfortunately, this scenario is becoming more common. The ISIS
- process group paradigm is being used to represent more and more
- lightweight objects in distributed applications. For example, the
- original design of the Deceit file system used an ISIS process group
- to represent a collection of servers maintaining replicas of a file.
- Each active file in the file system had its own corresponding process
- group. This initial design suffered performance problems due to the
- interaction of the ISIS protocols when communication switched between
- groups and when failures occurred. The author (Alex Siegel) then
- underwent a redesign of the system to reduce the number of process
- groups used in the system and substantially improved the performance.
-
- The lightweight process group mechanism in Horus will provide an
- interface similar to the process group interface present in ISIS
- today. However, the system will detach the interface from the
- communication protocols, allowing multiple lightweight process groups
- to share the same process group infrastructure. The desired effect
- being to retain the process group paradigm for use with lighter weight
- objects yet provide good performance, avoiding overhead due to
- competing process group interactions where possible. The system will
- also act as an environment for looking at different process group
- semantics, this aspect though is more interesting from a research
- point of view than from the user base. In addition the interface will
- allow for pipelining of distributed operations in a manner similar to
- the asynchronous I/O operations provided by several flavors of UNIX.
- This will retain the strong oredering guarantees provided by ISIS and
- allow for the clean notification of error conditions, and synchronous
- results.
-
- As Ken mentioned the LWG mechanism lies at the OS interface (the Horus
- kernel, MUTS, can be dropped into the OS's system space for efficiency)
- and will be used by the tools built on top of the system. It will
- provide the foundation for object oriented interfaces and tools built in
- the user level libraries.
-
- The following papers are relevant to lightweight process groups and
- the design of Horus.
-
- "Lightweight Process Groups" by Bradford Glade, Kenneth Birman,
- Robert Cooper, and Robbert van Renesse to appear in the proceedings
- for the OpenForum '92 Technical Conference in Utrecht, Holland in
- November.
-
- "Reliable Multicast between Microkernels" by Robbert van Renesse,
- Kenneth Birman, Robert Cooper, Bradford Glade, and Patrick
- Stephenson in the Proceedings of the USENIX workshop on
- Micro-Kernels and Other Kernel Architectures.
-
- "A RISC Approach to Process Groups" by Robbert van Renesse, Robert
- Cooper, Bradford Glade, and Patrick Stephenson in the Proceedings of
- the Fifth ACM SIGOPS European Workshop.
-
- These papers may be obtained via anonymous FTP from ftp.cs.cornell.edu
- (128.84.218.75). They can be found in the directory /pub/horus under
- the names lwg.ps, micro.ps, and sigops.ps respectively.
-
- I would like to hear from people who have ideas for new tools that
- they would like to see and who have requests for better support of
- object orientation, etc. This is your opportunity to shape the design
- of the user interfaces provided in this new generation of ISIS.
- Please feel free to discuss any aspects of the ISIS toolkit that
- really like, really hate, or would really like to see. Email to me
- can be sent to glade@cs.cornell.edu, or feel free to post here.
-
- Thanks, I look forward to your input.
-
- -Brad.
-
-
- --
- Bradford B. Glade | glade@cs.cornell.edu
- Dept. of Computer Science | Tel. (607) 255-9124 (Office)
- 314 Upson Hall Cornell Univ. | FAX (607) 255-4428 (CS Dept.)
- Ithaca, NY 14853 |
-