Mor Harchol-Balter

Research Interests

I am interested in the performance analysis and design of computer systems, particularly distributed systems. I work on finding analytical models which capture the important characteristics of a computer system and allow me to redesign the system to improve its performance.

I believe that many fundamental conventional wisdoms on which we base system designs are not well understood and sometimes false, leading to inferior designs. One problem is that many of our existing beliefs stem from analysis which is based on Markovian workloads (exponentially-distributed job sizes). However today's measured workloads show much greater variability in job sizes than previously assumed and furthermore show heavy tails. In light of current empirical measurements, my research challenges existing age-old heuristics. Here are just a few examples:

• Thousands of "load balancing" heuristics do exactly that -- they aim to balance the load among the existing hosts. But who said that's necessarily a good thing?
• Migration policies for networks of workstations and for distributed servers direct jobs to the host with least load. That seems good from the job's perspective, but is it best for the system overall?
• Given a choice between a single machine with speed s, or n identical machines each with speed s/n, which would you choose?
• Migrating active jobs is generally considered too expensive. Killing jobs midway through execution and restarting them from scratch later is even worse! Says who?
• Ever notice that the "proven best" scheduling policies like Shortest-Remaining-Processing-Time-First (SRPT) are never used in practice? There's a fear that the big jobs will starve. Is this true?

Problem areas I currently work in include: Web server design and implementation, queuing theory, scheduling and resource allocation, analysis under heavy-tailed workloads, routing and scheduling in networks.

In the News

Research Area

Computer Systems

Keywords

Scheduling of computer systems, workload characterization, heavy-tailed distributions

Education

PhD, 1996
Cognitive Science
University of California, Berkeley