In this paper we investigate the possibility to use neural networks to implement a new cache block placement strategy in cache memories. Based on the principle of space locality, we propose a cluster based cache block placement strategy more adapted to program behavior. The capacity of neural networks to produce such clusters is evaluated and cross-validated with standards statistical techniques. Positive results for the set of benchmarks used were obtained.

Towards self organizing cache memories using neural networks

Cache memories are essential components in all existing commercial microprocessors. In order to attain best performance, cache memories have to be managed either with hardware support or compiler support. The compiler approach makes use of specialized cache memory management instructions to generate an optimal management. This is done by generating an optimal scheduling of these specialized instructions for the program being compiled. Up to now, conservative approaches have been used to tackle this issue despite the occurrence of unpredictable real time events and the fact that many variables are imprecise. This explains the unstable performances of these algorithms varying from excellent to very poor. We propose to make use of a fuzzy scheduling approach to deal with the problem.

Fuzzy scheduling in compilers optimizations

We propose an approach for compressing address traces with access patterns irregularity. This approach is based on a cluster analysis of the trace. This analysis do not rely on any locality assumption and thus is suitable in cases where the space locality is poor due to access patterns irregularity. This approach can be used efficiently as a complement to other compressing techniques which rely exclusively on locality to achieve their results. As much as a 50% reduction in the storage requirements have been observed on synthetic traces.

Taking into account access patterns irregularity when compressing address traces

Working sets are a very useful notion in computer architecture to describe the set of memory locations used at a given instant of time in the computation. Working sets are closely related to the principles of spatial and temporal locality of references. Although these notion and principles are widely known by computer architects since a long time there is no widely accepted formal definition of what is a working set. This paper describes how with the help, of FCM the notion of fuzzy working sets match pretty well the ambiguous accepted formulation and can then be used for a fuzzy inference system for cache management.

http://ieeexplore.ieee.org/iel3/4033/11965/00551797.pdf

Cache memories, fuzzy localities and fuzzy working sets

In this paper, we propose to reduce execution time and to gain predictability by making use of a concurrent hardware software scheme for memory hierarchies. Making use of memory hierarchies will allow reducing memory access time while concurrency will relax memory bandwidth resource constraint. The software part of the scheme makes a static analysis of the real time application and generates a file containing special controller instructions. These instructions are generated and scheduled using artificial intelligence optimization techniques so to assure an optimal concurrent management scheme of the memory hierarchy. The hardware part is composed by specially designed memory controllers which are connected to a dedicated bus which allows access to all the memory hierarchy levels. These controllers will execute the instructions associated to the application concurrently with the execution of the application on the microprocessor. Bus contention is avoided between the microprocessor executing the real time application and the controllers on the dedicated bus due to the good scheduling generated at compile time. >

http://ieeexplore.ieee.org/iel2/1132/7852/00333088.pdf

O. Hammami

Papers

A concurrent hardware software management scheme for memory hierarchies