Lee Windsor Hoevel

Fractal nature of software-cache interaction

This paper uses fractals to model the clustering of cache misses. The clustering of cache misses can be quantified by a single number analog to a fractional dimension, and we are intrigued by the possibility that this number can be used as a measure of software complexity. The essential intuition is that cache misses are a direct reflection of changes in locality of reference, and that complex software requires more frequent (and larger) changes in this locality than simple software. The cluster dimension provides a measure (and perhaps the basis for a model) of the intrinsic differences between workloads. In this paper, we focus on cache miss activity as a discriminate between interactive and batch environments.

Measures of ideal execution architectures

This paper is a study in ideal computer architectures or program representations. We define measures of “ideal” architectures that are related to the higher-level representation used to describe a program at the source language level. Traditional machine architectures name operations and objects which are presumed to be present in the host machine: a memory space of certain size, ALU operations, etc. An ideal language-based architecture is based on a specific higher-level (source) language, and uses the operations in that language to describe transformations over objects in that language. The notion of ideal is necessarily constrained. The object program representation must be easily decompilable (i.e., the source is readily reconstructable). It is assumed that the source itself is a good representation for the original problem; thus any nonassignment operation present in the source program statement appears as a single instruction (operation) in the ideal representation. All named objects are defined with respect to the scope of definition of the source program. For simplicity of discussion, statistical behavior of the program and language is assumed to be unknown; Huffman codes are not used. From the above, canonic interpretive (CI) measures are developed. CI measures apply to both the space needed to represent a program and the time needed to interpret it. Example-based CI measures are evaluated for a variety of contemporary architectures, both host-and language-oriented, as well as a CI-derived language-oriented architecture.

ALAN: A (Circuit-Switched) Local Area Network

Our view is that people (that is, programmers and general computer users) tend to work together in small groups, where information and resources are shared freely within a group. Most interactions occur within the group (called a cluster)-sending messages, exchanging data, sharing a printer. Communication outside the cluster-sending mail to someone in another group or using a large number cruncher-is comparalively rare. Under this hypothesis, it is advantageous to optimize and simplify interactions within the group. This paper describes our attempt to design a cluster network based on a nonblocking crosspoint switch, which we call ALAN (a local area net). ALAN clusters are star-connected-with intelligent workstation nodes (PCs) at the points of the star, and the ALAN switch at the center.