Overview DataTAG task 3.1: Tools and Results

This page gives an overview of the tools and results that are related with task 3.1 defined in Work Package 3 from DataTAG. The following Web sites are available:

Workplan

• Before the work was started, a Workplan was created.
  Here we describe the steps that have to be taken to meet the objectives that were described in the workplan proposal task 3.1.

Measurement of layer 2 network characteristics

As a first step, layer 2 characteristics of long distance circuits were measured and tools were examined for testing and monitoring the network.

• Performance of the trans-Atlantic lambda with UDP.
  To get a feeling of the behaviour of a long distance (long delay) circuit setup with layer 2 equipment, performance measurements were done sending a single stream of UDP packets. This proved to be very worthwhile research, since it showed very significant difference in performances, based on the chosen settings.
• Network Test and Monitoring Tools.
  In this site an overview has been given of the proposed test and monitor tools to be used. Also short description of these tools have been given. Useful, non-default configurations are discussed and modifications and extensions of the discussed tools are available for downloading.

TCP and UDP characteristics of trans-Atlantic lambda's

Using the results above, complex tools were designed or extended, and thorughput two trans-Atlantic lambda's were measured. Both UDP and TCP characteristics were measured.

• Netherlight Lambda Test Results.
  In this site the monitor and test results are presented that were obtained at the Netherlight Lambda from SURFnet. Because both networks have about the same characteristics (long fat networks with limited routing and therefore relatively few memory buffer capacity, these results may serve very well as a testbed for the DataTAG connection from which some test results are also presented in the next item.
• DataTAG CERN - StarLight Link Test Results.
  In this site the test results are presented that were obtained at the DataTAG test link between CERN, Geneva and StarLight, Chicago.

Tuning of TCP parameters

In order to further study the performance of TCP traffic, tuning of TCP parameters was doing in collaboration with researchers in task 2.1.

• Kernel delay in TCP.
  TCP measurements with kernel delay. Performed in task 2.1 after TCP and UDP bulk transfers measurements showed a large discrepency.
• Transmit queue length variations in TCP.
  Measurement of the throughput, by varying the transmit queue length in TCP.

Ongoing work

Due to the bad performance of TCP on high bandwidth long delay circuits, as we have measured above, new protocols, intended to be used on dedicated circuits are designed. Work is ongoing to test the behaviour of these new protocols in combination with regular TCP traffic.

Preliminary results already exists, but is currently not ready for public publication. DataTAG member are advised to contact the AIR research group of the UvA for results.

Deliverables

Both deliverables of task 3.1 have been finished and published:

Deliverable 3.1

Evaluation, selection and adaptation of network performance & monitoring tools. (prototype in MS Word format, 667 kB).

This 13-page word document introduces the monitoring and test tools environment. This information can also be found at the website. The report also mentions some tools still under investigation, while the site does not. On the other hand, the site contains pointers to downloads, man pages and on-line examples, while the document does not.

Deliverable 3.2

Analysis of Performance Measurements. (report in MS Word format, 2.7 MB).

This 100-page report describes in detail the work done in task 3.1. Most of the information is also available in the sites overhead. For easy access, see the executive summary in HTML format.

Abstract: Basic performance measurements on transatlantic high speed lambdas has shown that user level data transfer are hampered by dependencies of the TCP protocol on the properties of the underlying network. Good throughputs can be obtained by making specific choices for Lambda bandwidth provisioning, sophisticated hard- and software, adaptations to the TCP protocol and by the use of multiple data streams. Monitoring tools are selected and presented to show the discrepancy between maximum attainable and application level achieved throughput or goodput.