Measurement procedure

In this section a description of the RTPL package ( Remote Throughput, Ping and Load) is given in more detail. As already explained in the introduction, this package does periodically net performance measurements between a set of hosts. The measurements are performed by a, so called, control host. This workstation starts the net performance measurements at each of the participating workstations with a, so called, remote shell command¹. Also the results of the measurements are send back via the remote shell command. This process is schemed in figure 1.

  

Figure: The control host starts the measurements at all hosts i by means of remote shells. The host i performs the measurements to the hosts j The results are send back by host i to the control host.

At each host i of the set hosts, described above, the following net performance measurements are executed:

Throughput.

Formal definition from RFC 1224 [2]: ``The maximum rate at which none of the offered frames are dropped by the device''. It is a way to quantify the traffic flow which can be handled by a network connection. Default it is measured for the connections from the current host i to all other hosts, but it is also possible to skip connections. The throughput is measured with the public domain command netperf.

Roundtrip or Ping.

This Internet application is described in RFC 2151 [3] by the paragraph quoted below:

Ping, reportedly an acronym for the Packet Internetwork Groper, is one of the most widely available tools bundled with TCP/IP software packages. Ping uses a series of Internet Control Message Protocol (ICMP) [4] Echo messages to determine if a remote host is active or inactive, and to determine the round-trip delay in communicating with it.

The roundtrip time quantifies the response offered by a network connection. It will be measured, before the throughput, across the same connections as the throughput. The roundtrip time is measured with the system command ping.

Load.

This is expressed here as # fully active processes at a host. It is no network quantity, but it may help to explain unexpected performance decreases. The load is measured at the current host i, using the system command uptime.

The sampling of the results at the control host and the measurements at all hosts, participating in the tests, are performed by scripts in the scripting language Perl. Perl is a powerful scripting language which is available for many platforms, including Unix, Windows NT/9Xand MacOS.

The Perl script at the control host collects the results of the measurements for each host i and stores the results in ZIP compressed data files. The ZIP compression is used to reduce disk space and download time. See subsection 2.2 for a description of the data files.

The tests are periodically started with the so called crontab² command. A sample crontab input file, which is specific for a hosts set, will be generated at the installation of that set. When the previous test is not ended when a new one is started by crontab, the new test terminates. However, to prevent deadlocks, after some terminations the new version kills the old one and continues. Timeouts are used in the test scripts to prevent deadlocks.