Cluster:LLK Poster Structure

Abstract

Network latency continues to be a bottleneck for many computational science applications on Beowulf clusters. Using open source tools, we develop technology for making nanosecond precision measurements between timing points within the 2.6 Linux kernel. Using this technology, in conjunction with low-level and application benchmarks, we establish that there is significant network latency in the kernel and identify its origin. Lastly, we survey software-based latency reduction technologies and make recommendations based on our measurements. (original)

Keywords:

The Problem

Survey/Current State of Affairs

• Light, detail in #Measurement Techniques.
• Include dead ends here.
• List of approaches.

Places to look:

• Beowulf archives
• ACM
• IEEE
• Citeseer
• Google Scholar

Design (of the Experiment, Methodology)

Measurement (Techniques)

• TCP/IP stack time point placement
• tp_timer
• high res timing w/counters

(Analysis) Tools

• /proc -> database
• PHP/gnuplot visualization

Benchmarks

• netpipe (low level)
• MrBayes (high level)

Results (of the Experiment)

(tabular, incl. test environment)