RD11: IBM & HP TCP/IP and direct channel speed measurements

Overview of FCS equipment under test

CERN's equipment bought from industry is now all running at version 4.3:

The most up-to-date results of all measurements done on this heterogeneous equipment, can be found in the document "Fibre Channel Equipment under Test". In this document the latest configuration of CERN's FCS equipment and a summary of the latency and throughput measurements of the different workstations can be found. The document is available in two forms:

The document "Fibre Channel Performances with IBM Equipment" describes tests performed at IBM-ECSEC in Rome, that measured communication performances at the user level when exchanging data between IBM workstations via an eight-port IBM Fibre Channel fabric. This document contains many graphs of overhead, latency, throughput and processor usage. It also describes how Fibre Channel can be used in High Energy Physics and Low Energy Physics applications.


Old documents


IBM RS/6000

Within the framework of RD11 (P12) work has been done at CERN to measure communication times for transferring data from memory to memory between RS/6000 IBM workstations via a 16 port Ancor Fibre Channel Fabric (or Switch). Both TCP/IP and Direct Channel measurements are done.

The paper "Communication benchmarks with the Ancor Fibre Channel Fabric" describes in detail those measurements. The document is available in two forms:

HP9000/700

Similar measurements are made with Hewlett Packard 9000/700 workstations with Fibre Channel interfaces. Also measurements between IBM and HP workstations are made.

The paper "Communication benchmarks with HPs and IBMs connected around an Ancor Fibre Channel Fabric" describes in detail those measurements. The document is available in two forms:

High Speed TCP/IP (RFC 1323)

After the above two papers were written, measurements were made in November 1994 with High Performance TCP/IP running on the machine's kernels. Today, most workstations (IBM, HP, DEC, SGI at least) support this potentiality, but to our knowledge, none implements it by default. It is the user's responsibility to switch it on. Although a considerable improvement can be expected as we have measured on both HP and IBM machines.

The paper "High Performance TCP/IP communications with HPs and IBMs connected around an Ancor Fibre Channel Fabric" describes in detail the measurements with the High Performance TCP/IP. The document is available in two forms:

Contact Fabrice Chantemargue (chantemargue@cern.ch), Tel +41 22 767 7759 for more details on this project.


This is one of the CERN High Speed Interconnect pages - 9 June 1995 - Erik van der Bij