clarke@csri.toronto.edu (Jim Clarke) (02/28/89)
(SF = Sandford Fleming Building, 10 King's College Road)
(GB = Galbraith Building, 35 St. George Street)
SUMMARY:
SYSTEMS SEMINAR - Mon., March 6, 2 p.m. in SF 4103 -- Robert C.B. Cooper
"Hierarchical Process Groups and Large Scale Broadcasts"
SYSTEMS SEMINAR - Tues., March 7, 2 p.m. in GB 244 -- Jim Donahue
"The Modula-3 Story"
INFORMATION SYSTEMS SEMINAR - Thurs., March 9, 2 p.m. in SF 1013
-- Alexander Tuzhilin
"Using Relational Discrete Event Systems and Models for
Prediction of Future Behavior of Databases"
---------------
SYSTEMS SEMINAR - Monday, March 6, 2 p.m. in Room SF 4103
Robert C.B. Cooper
Cornell University
"Hierarchical Process Groups and Large Scale Broadcasts"
Process groups are a useful abstraction for reliable broadcast-based dis-
tributed systems, and have been implemented in systems such as V and ISIS.
But as process groups grow large, most broadcast algorithms become unrea-
sonably inefficient.
In this talk I will present a hierarchical structure for process groups.
Such an organization not only supports more efficient tree structured
broadcasts, but actually obviates the need to broadcast to the entire
membership of a large group for most common communication patterns.
This approach is being investigated within the ISIS distributed programming
system. The aim is to extend ISIS to support groups comprising hundreds of
members on a cluster of local networks. A secondary goal of this research
is to provide compatibility with the existing ISIS toolkit, permitting ex-
isting programs, with minor changes, to take advantage of hierarchical pro-
cess groups.
SYSTEMS SEMINAR - Tuesday, March 7, 2 p.m. in Room GB 244
Jim Donahue
Olivetti Research Center
"The Modula-3 Story"
In 1987, a joint project was started between the Olivetti Research Center
and the Digital Equipment Corporation Systems Research Center to design a
new systems programming language. The resulting language, Modula-3, is de-
fined by the Modula-3 Report, which will be jointly published by ORC and
DEC SRC, and is currently being implemented by ORC. Modula-3 descends from
Mesa, Modula-2, and Modula-2+; it also resembles its cousins Object Pascal,
Oberon, and Euclid.
In this talk, I will give the historical background of the Modula-3 pro-
ject, describe the principles that drove the language design, and discuss
the most important features of the language. I will focus on the type sys-
tem, since that is the most significant change from Modula-2 and Modula-2+.
I will also discuss briefly the Olivetti Modula-3 implementation effort.
INFORMATION SYSTEMS SEMINAR SPONSORED JOINTLY BY DCS AND FLIS
Thursday, March 9, 2 p.m. in Room SF 1013
Alexander Tuzhilin
New York University
"Using Relational Discrete Event Systems and Models for
Prediction of Future Behavior of Databases"
The problem of prediction of future behavior of relational databases is ad-
dressed in this presentation. In the first part of the presentation, for-
malisms to model potential future behavior of a database, based on the con-
cepts of Relational Discrete Event Systems and Models (RDESes and RDEMs),
are defined. An RDES describes all possible non-deterministic future
behaviors of a database system in terms of finite and infinite trajectories
over the consistent states of the database. An RDEM is a finite formal
description of an RDES. Two specific non-deterministic RDEMs, Production
System (PS) and Recurrence Equation (RE) RDEMs, are considered and formally
compared in terms of their expressive powers. Also, various PS RDEM inter-
preters are analyzed and formally compared with respect to several meas-
ures. A specific interpreter, the parallel deterministic interpreter, is
determined to be better than others in many respects. It is also esta-
blished that PS RDEM with this interpreter has more expressive power than
RE RDEM. Therefore, PS RDEM with the parallel deterministic interpreter is
chosen as the canonical formalism to model future behavior of relational
databases. In the second part of the presentation, Predictive Query
Language, based on temporal logic, is defined. Future predictions are ob-
tained as answers to queries expressed in this language. Implementation of
production system RDEM and Predictive Query Language is also discussed.
The suitability of various control strategies to restrict non-determinism
and improve system's performance is under study.
--
Jim Clarke -- Dept. of Computer Science, Univ. of Toronto, Canada M5S 1A4
(416) 978-4058
clarke@csri.toronto.edu or clarke@csri.utoronto.ca
or ...!{uunet, pyramid, watmath, ubc-cs}!utai!utcsri!clarke