Research

Elizabeth Varki
Department of Computer Science
University of New Hampshire
Kingsbury Hall
Durham, NH 03824

Phone: (603) 862-2319
Email: varki@cs.unh.edu
www: www.cs.unh.edu/~varki

My area of expertise is performance evaluation. I evaluate the performance of computer and communication systems, storage devices, and software systems. I am a modeler: real-world is complex and it is difficult to understand what features of the real-world drives performance. For example, when you click a link on your web browser, what determines the speed with which the page is downloaded on your computer - the speed of your computer, or the network bandwidth, or the speed of the web server, or some other factor? This leads to further questions: what determines the speed of your computer - the CPU speed or the cache size or the memory bus speed - leading to futher evaluation at lower levels. There are many, many parameters in a real system, which makes it difficult to see the forest for the trees. I determine what parameters are essential to determining the performance of the real-world system under consideration. I then develop a mathematical model that incoporates only those parameters that drive the performance. Thus, my model is an abstraction of the real system that incoporporates the essential parameters that drive performance. This model can then be used to understance the behavior of the real system.

I develop performance models which compute mean performance measures such as response time and throughput. These are articulated by queueing models. When computing performance of distributed systems - computers around the globe connected by the Internet - I use graph models. Once I develop the model, I use existing algorithms, to compute performance measures from the model. If algorithms/tools to compute mean performance measures from the model do not exist, then I develop (or try to develop) the algorithm.

I develop new technology if existing technology is not keeping pace with improvements in hardware and software. For example, my student and I have developed RAIDX for arrays of dissimilar disks since RAID was developed for arrays of similar disks.

Listed below are my papers, classified by topic, starting from my most recent research interests. It is ironic that my best papers on most topics haven't been published.

RAIDX: RAID for heterogenous array of disks

We propose a new organization for organizing an array of disks of different types, sizes, and speeds. RAIDX does not stripe data across disks, it bundles chunks of data across disks.

GPSclock: A performance model to track time and flow capacity of end networks

A graph model that models time at end networks. The time determines the available bandwidth capacity, so this graph models flow capacity at end networks. This model underlies maximal flow problems between end networks.

Prefetch cache replacement technique

Prefetch cache

Arrival queues

Fibre channel

RAID

Fork-Join queues

Return to my Home Page