WSC 2005

WSC 2005 Final Abstracts

General Applications C Track

Wednesday 10:30:00 AM 12:00:00 PM
General Applications: Computer Networks

Chair: Nava Aizikowitz (IBM Haifa Research Lab)

Digital Video Cluster Simulation
Martin N. Milkovits (SeaChange International)

The advent of Video On Demand (VOD) services is made possible by specialized high performance, high reliability computer systems. These systems must maintain the constant bandwidth required by video and guarantee fault resiliency. SeaChanges digital video system meets the VOD challenges with a scalable cluster solution using a patented data distribution algorithm for efficient data redundancy. The SeaChange digital video cluster incorporates several different bus and fabric technologies to deliver high performance and data reliability. But, the existence of many data paths and congestion scenarios make it difficult to determine the maximum performance of, or bottlenecks in, the system. Therefore, a simulation model to represent the internal fabric of a VOD cluster from SeaChange is designed, implemented and verified.

Realistic Internet Traffic Simulation Through Mixture Modeling and a Case Study
Song Luo (University of Central Florida) and Gerald A. Marin (Florida Institute of Technology)

Internet background traffic modeling and simulation is the main challenge when constructing a test environment for network intrusion detection experiments. However, a realistic simulation of network traffic through analytical models is difficult, because the classic distributions are usually ineffective when applied to traffic-related random variables. A modeling and simulation approach using heavy-tailed mixture distributions is introduced in this paper. In the case study, this approach is used to build analytical models for random variables of several major Internet applications (FTP, HTTP, SMTP, POP3, SSH) of a campus network. Several statistical features of an NS2 simulation are compared against those of the traffic traces being simulated. The comparison indicates that the simulation is statistically similar to the real traffic.

Component-Based Performance Modeling of a Storage Area Network
Nava Aizikowitz, Alex Glikson, Ariel Landau, Bilha Mendelson, and Tommy Sandbank (IBM Haifa Research Lab)

This work explores performance issues of system-level interactions by means of performance modeling. We focus on I/O performance in a storage area network (SAN), namely, the performance of I/O interactions of host servers and storage subsystems via the SAN fabric. We present a component-based simulation performance model, which supports a rich variety of both existing and future storage subsystems, allows for some basic network configurations, and addresses the major I/O aspects of the server operating system. The model's flexibility allows for easy parameter modifications, configuration adjustments, architecture manipulations, and experimentation. The experiments presented in this paper demonstrate some of the ways this model can be utilized, such as data placement, I/O manipulation, and the evaluation of execution alternatives, and shows the types of performance insights that may be gained.