Performance Analysis of Available Bandwidth Estimation Tools for Grid Networks

Abstract

Modern large-scale grid computing for processing advanced science and engineering applications relies on geographically distributed clusters. In such highly distributed environments, estimating the available bandwidth between clusters is a key issue for task scheduling. We analyze the performance of two well known available bandwidth estimation tools, pathload and abget, with the aim of using them in grid environments. Our experiments consider the accuracy of the estimation, the convergence time, their level of intrusion in the grid links, and the ability to handle multiple simultaneous estimations. Overall, pathload represents a good solution to estimate available bandwidth in grid environments.

Refining the estimation of the available bandwidth in inter-cloud links for task scheduling

“…Even if the available bandwidth could be precisely estimated by existing tools [5], such estimation would not represent the fluctuation of the available bandwidth during the execution of the workflow, which can take from seconds to days. Overestimating the available bandwidth yields to longer data transfer times than expected, enlarging makespans which can potentially surpass the established deadline, in addition to increase the monetary cost of the workflow executions as a consequence of longer rental periods.…”

“…Information on the available bandwidth in inter-clouds communication channels is of paramount importance to schedule workflows on hybrid clouds, since the available bandwidth directly impacts the duration of data transfer between clouds which can potentially increase the makespan as well as the monetary cost to execute the application [4]. Since inter-cloud communication channels are usually Internet links, the available bandwidth in these links is quite variable [5]. Mechanisms that deal with such variabilities as well as with imprecise estimations of the available bandwidth need to attempt to diminish the negative impact of misleading information on the production of schedules.…”

“…Such relationship would not be possible to assess by just forecasting the available bandwidth variation or measuring the error of available bandwidth estimation [5], since the effect of both on different workflows differs due to distinct structure of task dependencies. Therefore, the multiple linear regression procedure consider past makespan estimations made for a workflow as well as the errors associated with these estimations.…”

“…However, such assumption is not always true. Actually, it was shown in [5] that tools for estimating available bandwidth produce estimations with large variability. Especially in hybrid clouds, schedules can be negatively affected by misleading bandwidth estimations during the execution of workflows.…”

Performance Analysis of a DEKF for Available Bandwidth Measurement

“…Its estimate is used for route selection, quality of service (QoS) verification, and traffic engineering. More recently, AB estimation has increased its importance with respect to new demanding scenarios like cloud computing, voice-over-IP (VoIP), and multimedia networks, where support tools for traffic shaping and traffic policing as well as developing network management strategies are crucial [2][3][4][5][6][7]. Moreover, available bandwidth can directly affect the performance of distributed systems and clocks [8,9].…”