CGL Meeting Agenda

Wednesday, August 6th, 1997

Location:
DC 1304
Time:
13:30
Chair:
Ed Dengler

1. Adoption of the Agenda - additions or deletions

2. Coffee Hour

Coffee hour this week:
???
Coffee hour next week:
???

3. Next meeting

Date:
August 13th, 1997
Location:
DC 1304
Time:
13:30
Chair:
Petra Englemann
Technical presentation:
Ed Dengler

4. Forthcoming

Chairs:

  1. Dave Evans (8/20)
  2. Glenn Evans (8/27)
  3. Patrick Gilhuly (9/03)

Tech Presenters:

  1. Petra Englemann (8/20)
  2. Dave Evans (8/27)
  3. Glenn Evans (9/03)

5. Technical Presentation

Presenter:
William Cowan
Title:
The Economics of Just-in-Time Compilation
Abstract:
The combination of late-binding and deep pipelining makes static optimization of object-oriented code difficult. Just-in-time compilation makes it possible to compile and/or optimize at run-time, using observed properties of the running code. The decision whether to compile or to interpret then becomes an economic one. This short talk will discuss the issues and describe a formalism in which execution can be optimized. (Warning. Java will be mentioned.)

6. General Discussion Items

7. Action List

8. Director's Meeting

9. Seminars

WHAT:  Master's Thesis Presentation
WHO:   Raj Rathee
WHEN:  Thursday, August 14, 1997, 10:00-11:00 a.m.
WHERE: DC 1331

``Using Advance Knowledge of Transactions to Provide Efficient Recovery
Management''.

ABSTRACT:
Certain  classes  of  systems,  such  as  some embedded
control  systems, have the property that their database
schema, as well as the set of transactions, is known in
advance.  However,  transaction support in such systems
is  provided  by methods that were designed for systems
where  this information is not available. Consequently,
these methods fail to exploit it.

This thesis presents a recovery algorithm ( Fuzzy_PBL )
that  takes  advantage of known transaction data access
patterns  to  reduce  recovery  related  overhead.  The
Fuzzy_PBL  algorithm utilizes transaction level logical
logging  and  fuzzy  checkpointing. Logical transaction
logging  minimizes  the number of log entries that must
be   written.    Fuzzy   checkpointing  causes  minimal
disruption   to   normal  transactional  activity.  The
combination  of  these  two  techniques can potentially
lead  to  unrecoverable  database  states  in case of a
system  failure.   The Fuzzy_PBL algorithm utilizes its
advance  knowledge  of  transaction  access patterns to
avoid this problem.

A  performance  analysis  comparing  Fuzzy_PBL  with  a
traditional  recovery method is presented. Results show
that  the Fuzzy_PBL approach pays off through increased
transaction throughput.

10. Lab Cleanup (until 14:30 or 5 minutes)