Static Memory Management for Logic Programming Languages

1
Static Memory Management for Logic Programming
Languages

• Quan Phan
• DTAI Analysis group
• Katholieke Universiteit Leuven, Belgium
• Doctoral Consortium, ICLP 2006

2
Outline

• Motivation
• Memory management in logic programming
• Literature on static memory management
• My research
• Research goal
• Current status
• Expected achievements

3
Memory management in logic programming

• Logic programming is declarative
• Focus on WHAT, not HOW
• No malloc/free
• Automatic management of the heap
• Instant reclaiming
• Rely on backtracking
• Not enough deterministicshallow backtracking
• Runtime garbage collection
• Effectively recover, gt 90 of the garbage
• Problem runtime overhead, time-unpredictable

4
1. Region-based memory management (RBMM)
Basic idea
The heap

• Heap regions

X
.

• Terms distributed over regions

.

1

• Release the whole region at once

2
Y
.
3

.
Z
.
.
5
1. RBMM Practice

• The compiler
• decides about regions
• and region lifetime
• adds creation/removal instructions

6
1. RBMM

• An extensive research field in functional
  programming
• Mostly SML
• Region inference is type-based analysis
• More recently C, Cyclone, Java
• For logic programming
• application to WAM-based Prolog by Henning
  Makholm Kostis Sagonas
• Focus runtime support for nondeterminism.
• Borrow region inference from SML

7
2. Compile-time garbage collection (CTGC)

• Idea instead of releasing, reuse allocated
  memory cells for new terms
• Practice
• CTGC for Mercury language
• Structure sharing analysis, liveness analysis ?
  reusable cells
• Reuse locally inside procedure

8
2. CTGC

• append(X, Y, Z) -
• (
• X ,
• Z Y
• X gt Xe Xs,
• append(Xs,Y,Zs),
• Z lt Xe Zs
• ).

X
.
1
1
.
.
2
2
Z
.
.

Y
.
3

9
Advantages of static approach

• Performance
• Reasoning is done at compile-time
• Competitive runtime no or less RTGC needed
• Always better performance? Non conclusive, due to
  locality, runtime overhead of new instructions,
  region management, but at least more
  time-predictable
• Competitive memory consumption
• Reuse or reclaim all the garbage memory
• Let programs run in linear space

10
Research goal

• Study the possibility and practicality of a
  hybrid memory management system that combines
  RBMM and CTGC.
• How to combine?
• Memory management is both theoretic and
  experimental
• build and test systems in practice
• a good thing, for which programs?

11
Current status First step

• RBMM Analysis for Mercury
• same context with the only CTGC system
• The algorithm three phases
• Region points-to analysis build the memory model
• Region liveness analysis decide region lifetime
• Source-to-source transformation add creation and
  removal instructions

12
Current status First step

• More information
• Quan Phan and Gerda Janssens. Towards
  Region-based Memory Management for Mercury
  programs. Proceedings of the Colloquium on
  Implementation of Constraint Logic Programming
  Systems, CICLOPS 2006, Seattle, USA, Aug. 2006.
• Presentation at CICLOPS, 15h, Monday 21/8, in
  ???.

13
Ongoing work

• Modular region analysis
• Module system is critical for software
  engineering
• A challenge
• Region analysis is global
• Research is little on this problem, only one work
  exists
• Idea make the analysis local, allow callers to
  control callees behaviour.
• More precise region points-to analysis
• Study the hybrid system
• The operation of CTGC when memory is organized in
  terms of regions
• The cooperation among RBMM, CTGC and RTGC.

14
Expectation

• A programming system
• better memory footprint,
• runtime overhead on memory task is reduced
• flexible static memory management strategy
• which strategy performs best for what kind of
  programs.
• which combination often works effectively