b-jones@ttidca.TTI.COM (Bill Jones) (11/18/89)
Can someone out there PLEASE direct me to some literature (books, articles, ANYTHING) that describes ISIS? I'm interested in evaluating it for use in OLTP applications, but I can't seem to find any detailed info on it.
ken@gvax.cs.cornell.edu (Ken Birman) (11/21/89)
In article <7798@ttidca.TTI.COM> b-jones@ttidca.tti.com (Bill Jones) writes: > >Can someone out there PLEASE direct me to some literature (books, articles, >ANYTHING) that describes ISIS? I'm interested in evaluating it for use in >OLTP applications, but I can't seem to find any detailed info on it. Sure. First, let me mention that a complete publications list is maintained by my secretary, as well as a technical report distribution list. If you want to get a copy of the former or have your name added to the latter, drop a note to her (croft@cs.cornell.edu). Be sure to make it clear what you are asking for. We also distribute free reprints of our papers. I have asked Keith Marzullo to post something on META, so I won't do so here. If you want to read something general about ISIS we suggest that you start with one of the following: 1. The ISIS Distributed Programming Toolkit and the META Distributed System SUN Technology 2, 1 (Summer 1989) ... this gives an overview of the whole project in moderately technical terms, aimed at a systems administrator or programmer. 2. Chapters 13, 14 and the conclusions in: Distributed Systems. Sape J. Mullender, ed. Addison-Wesley, 1989. ISBN 0-201-41660-3. ... two chapters on ISIS, virtual synchrony, but also on other technologies in the same area (reliable broadcast, replication, etc.) Conclusion chapter asks "how robust" one can make a distributed system. 3. Exploiting Virtual Synchrony in Distributed Systems. Birman and Joseph. 11th ACM Symposium on Operating Systems Principles, December 1987. Also appearing as: Operating Systems Review, 22, 1, Dec. 1987, 123-38. ... the original paper on "virtual synchrony" and on the toolkit. 4. Reliable Communication in the Presence of Failures. Birman and Joseph. ACM Transactions on Computer Systems, \fB5\fP, 1, February 1987, 47-76. ... technical detail on how we implemented some aspects of ISIS; the system has now evolved quite far from this, however, so don't take it too literally. We also have a detailed programmer's manual, available for $15 from our group or for free in DVI format, as well as an extensive set of online manual pages for the system. Several papers are being written at present: One on distributed application management. One on partitioned applications One on making the ISIS multicast facilities faster. A paper on the Deceit file system architecture. A paper on scaling ISIS up to large settings. There are also several PhD dissertations from our group: T. Joseph (Low cost management of replicated data) F. Schmuck (The use of efficient broadcast protocols in async. dist. systems) K. Kane (Log based recovery in async. dist. systems) Again, reprints of these are available on request. Keith Marzullo, as mentioned above, also has several papers that might interest people reading up on this general area. His work focuses on sensors and realtime issues but makes use of ISIS as part of its implementation in the "meta" package. Anyone on our mailing list will get copies of these as they come out; please don't request early preprints as we are unable to accomodate that sort of thing. Also, although we are happy to provide reprints, we hesitate to provide on-line copy of our papers. This is because the figures are usually missing in on-line distributions and our papers tend to rely heavily on figures. Now, let me just comment that whereas ISIS might play a useful role in OLTP settings, ISIS per-se is definitely _not_ an OLTP technology. OLTP systems focus primarily on transactional access to files and objects. An early version of ISIS had this emphasis (see Proc. 10th SOSP) but we rejected this approach early in our effort. ISIS focuses more on the problem of consistent or coordinated actions in distributed services than on persistent data or serializability. On the other hand, we strongly believe that these are orthogonal problems, and that most OLTP systems will need a layer that does what ISIS does as well as having a layer that does what, say, CAMELOT does (an OLTP system developed at CMU). Our approach, in a nutshell, has been to use ISIS to glue together and control programs doing other things; OLTP servers and clients represent a good candidate for doing this. I wish I could tell you about what we are doing now in this area (control and "glue"), but I think it might be best to hold off until we are ready to publish and release some software.... Ken Birman