A Realistic Integrated Model of Parallel System Workloads

Abstract

Abstract-Performance evaluation is a significant step in the study of scheduling algorithms in large-scale parallel systems ranging from supercomputers to clusters and grids. One of the key factors that have a strong effect on the evaluation results is the workloads (or traces) used in experiments. In practice, several researchers use unrealistic synthetic workloads in their scheduling evaluations because they lack models that can help generate realistic synthetic workloads. In this paper we propose a full model to capture the following characteristics of real parallel system workloads: 1) long range dependence in the job arrival process, 2) temporal and spatial burstiness, 3) bag-oftasks behaviour, and 4) correlation between the runtime and the number of processors. Validation of our model with real traces shows that our model not only captures the above characteristics but also fits the marginal distributions well. In addition, we also present an approach to quantify burstiness in a job arrival process (temporal) as well as burstiness in the load of a trace (spatial).

Towards Non-Stationary Grid Models

“…However, more concepts (realized as sub-model parameters) increase the risk of an arbitrary choice inspired by the considered samples. Specifically, Minh et al [34] validate the generative model only in a narrow sense: it checks the coincidence of the previously described four concepts as fitted from experimental data with those fitted from a generated realization. In other words, if e.g.…”

“…Minh et al advocates for this goal [34], and specifically presents an integrated model of parallel bag-of-tasks (BoT) workload. The proposed model is built on four a-priori concepts (longrange dependence, temporal and spatial burstiness, BoT clustering, and correlation between parallelism and runtime), which implicitly define an ontology of parallel workloads (e.g.…”

Stochastic Tail-Phase Optimization for Bag-of-Tasks Execution in Clouds

“…It has been shown [3] that some bags of tasks have a skewed distribution, bounded by some maximum value. To model such bags, we generate workloads according to a truncated Levy distribution with a scaling factor (τ ) of 12 minutes and a maximum value (b) of 45 minutes, as shown in Fig.…”

Measuring Cloud Workload Burstiness

“…The slope can then be used as an indicator for workload burstiness. If there is a change in the workload burstiness and the plot is non-linear, then this method does not work [21].…”

“…2) Normalized entropy: Minh et al [21] introduce the use of normalized entropy as a measure of workload burstiness. It is known that the maximum value achievable for H is log N [11].…”