How to parameterize models with bursty workloads

Abstract

Although recent advances in theory indicate that burstiness in the service time process can be handled effectively by queueing models (e.g., MAP queueing networks [2]), there is a lack of understanding and of practical results on how to perform model parameterization, especially when this model parameterization must be derived from limited coarse measurements as is often encountered in practice. We propose a new modeling methodology based on the index of dispersion of the service process at a server, which is inferred by observing the number of completions within the concatenated busy periods of that server. The index of dispersion together with other measurements that reflect the "estimated" mean and the 95th percentile of service times are used to derive a MAP process that captures well burstiness and variability of the true service process, despite inevitable inaccuracies that result from inexact measurements. Detailed experimentation on a TPC-W testbed where all measurements are obtained via a commercially available tool, the HP (Mercury) Diagnostics, shows that the proposed technique offers a simple yet powerful solution to the difficult problem of inferring accurate descriptors of the service time process from coarse measurements. Experimental and model prediction results are in excellent agreement and argue strongly for the effectiveness of the proposed methodology under bursty or simply variable workloads.

Online Spike Detection in Cloud Workloads

“…[7] introduces the concept of characterizing burstiness in time series of workload arrivals using the index of dispersion. In [8], [9] the authors describe how the index of dispersion can be used to model and parametrize bursty workloads when investigating service times in multi-tier applications. [10] presents Fasttrack, a dynamic resource provisioning solution for multi-tiered applications that estimates the index of dispersion and utilizes it for determining when the workload is entering and exiting a bursty state.…”

Defragmenting the cloud using demand-based resource allocation

“…Casale et al [7] showed how to incorporate burstiness into the analytical queuing network models. These approaches provide sound foundational basis for medium term or offline capacity estimation, in contrast to our adaptive approach based on measured performance.…”

A Realistic Integrated Model of Parallel System Workloads

“…Temporal burstiness is the tendency of job arrivals to occur in clusters, or bursts, separated by long periods of relatively few or no arrivals [10], [19]. In fact, there always exist bursts in real workloads due to the occurrence of bags-of-tasks and idle periods during nights, weekends, holidays, etc.…”