Exploiting process lifetime distributions for dynamic load balancing

Abstract

We consider policies for CPU load balancing in networks of workstations. We address the question of whether preemptive migration (migrating active processes) is necessary, or whether remote execution (migrating processes only at the time of birth) is sufficient for load balancing. We show that resolving this issue is strongly tied to understanding the process lifetime distribution. Our measurements indicate that the distribution of lifetimes for a UNIX process is Pareto (heavy-tailed), with a consistent functional form over a variety of workloads. We show how to apply this distribution to derive a preemptive migration policy that requires no hand-tuned parameters. We used a trace-driven simulation to show that our preemptive migration strategy is far more effective than remote execution, even when the memory transfer cost is high.

Workload Modeling for Computer Systems Performance Evaluation

“…Migrating such processes from overloaded workstations to underloaded ones therefore will have a long-term beneficial effect on the load conditions. Moreover, it is easy to identify the long processes due to the conditional expectation of heavy tails: they are the processes that have run the longest so far [319,434]. Moving a randomly chosen process will only have a minor effect, because such a process will probably terminate shortly after being moved.…”

“…In fact, they are heavy-tailed [434,319]. Specifically, the probability that a process run for more than τ time has been found to decay polynomially rather than exponentially:…”

“…If we assume that process runtimes are exponentially distributed, this becomes impossible. Luckily, the real distribution tends to be heavy-tailed, so long-lived processes can indeed be identified leading to large benefits from migration [434,319]. Another example comes from garbage collection.…”

“…The dataset is the runtimes of Unix processes from 1994 [319]. These have a heavy tail, and figure prominently in Chapter 5.…”

“…The left plot is a dataset from 1994 used by Harchol-Balter and Downey, with about 185,000 processes that arrived over eight hours [319]. The right-hand dataset contains more than 450,000 Unix processes from a departmental server, covering about a month in 2005.…”