Organization

The topics can be classified according to different criteria.

In a general parallel architecture we find several memory layers to which we can apply HM techniques
(1) processor cache, (2) local memory (physically shared), (3) remote memory (either virtually shared, or accessed by message passing), (4) local disk device(s), and (5) remote disks.

There are other distinctions we can make

• From the point of view of parallelism expression and related overheads, there is a trade-off between threads and processes, and the use of shared memory versus that of explicit message passing. The two approaches are mixed when working with SMP Clusters.
• Are the access methods and caching policies fixed by the hardware, or are they user-definable?
• Some tools allow the user to define parallel HM computation, other tools implicitly exploit parallelism to improve HM access (PFS).

We have agreed on a first organization of the matter, which has turned into a set of 4 chapters

1. SMP cluster techniques -- (Martin)
2. hierarchical parallel computation and libraries to exploit it -- ``Hierarchical computation models and software tools'' (Massimo)
3. file systems -- ``File Systems: state of the art'' (Florin)
4. storage managing techniques -- ``Algorithmic and implementation of storage networks'' (Kay)

Within this general scheme, I outline my contribute in the following section.