Taming Wild Horses: The Need for Virtual Time-Based Scheduling of VMs in Network Simulations

The next generation of scalable network simulators employ virtual machines (VMs) to act as high-fidelity models of traffic producer/consumer nodes in simulated networks. However, network simulations could be inaccurate if VMs are not scheduled according to virtual time, especially when many VMs are...

Full description

Saved in:
Bibliographic Details
Main Authors: Yoginath, S. B., Perumalla, K. S., Henz, B. J.
Format: Conference Proceeding
Language:English
Subjects:
Ad hoc networks
Ad-hoc Wireless
Benchmark testing
Computational modeling
High-fidelity
Measurement
MPI
Multi-core
Multiplexing
Network Simulation
Processor scheduling
Time Synchronization
Virtual machine monitors
Virtual Machines
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The next generation of scalable network simulators employ virtual machines (VMs) to act as high-fidelity models of traffic producer/consumer nodes in simulated networks. However, network simulations could be inaccurate if VMs are not scheduled according to virtual time, especially when many VMs are hosted per simulator core in a multi-core simulator environment. Since VMs are by default free-running, on the outset, it is not clear if, and to what extent, their untamed execution affects the results in simulated scenarios. Here, we provide the first quantitative basis for establishing the need for generalized virtual time scheduling of VMs in network simulators, based on an actual prototyped implementations. To exercise breadth, our system is tested with disparate applications: (a) a set of message passing parallel programs, (b) a computer worm propagation phenomenon, and (c) a mobile ad-hoc wireless network simulation. We define and use error metrics and benchmarks in scaled tests to empirically report the poor match of traditional, fairness-based VM scheduling to VM-based network simulation, and also clearly show the better performance of our simulation-specific scheduler, with up to 64 VMs hosted on a 12-core simulator node.
ISSN:1526-7539
2375-0227
DOI:10.1109/MASCOTS.2012.18