Most progress has been made with the Active Measurement Program, where many more machines had been deployed, and functionality added.
In addition the Cichlid visualization engine has now been ported to the Windows 95/98/NT environment, and efforts are underway for major revision in a version 2.
- AMP machines
- the development of software to automatically detect significant changes in the network state and to inform interested personnel
- to identify where on a path that is exhibiting high latency or low throughput the problem is occurring.
- Sample AMP web output
- IPMP
- Simulations
The most significant AMP development over this quarter has been the deployment of a large number of monitoring machines. By the end of the quarter there were more than 30 monitors operational with more requests arriving each week. We expect over the next quarter to increase the number of monitors even further, perhaps to as many as 70.
Because of the scale of the deployment some problems have been encountered. In some cases these have highlighted network problems and improvements have been made by the sites concerned. In other cases local policies have made measurement difficult. In all cases technical staff at the deployment sites have been very supportive and have worked with NLANR staff to find workable solutions.
Further development of the AMP analysis and management software has been undertaken to reduce the day to day burden of the administration of the monitors and to support further growth of the number in monitors. This includes the development of "active mirroring" software that minimizes the data transfer burden required to send the measurement results to San Diego while providing at most a 5 minute lag in the availability of the data. This software may be useful to other projects has been made publicly available, along with documentation, at http://moat.nlanr.net/~tonym/mirror.
It has also been necessary to develop a status overview page (see http://moat.nlanr.net/Active/cgi-bin/status.cgi), which gives a one page summary of the status of all the monitors and the testing of the paths between them. This page has allowed us to more quickly discover and correct faults that stop the collection of data. A similar email system, which generates email when faults occur, has also been implemented.
It is expected that, as the number of monitors grows, the importance of administrative tools like these will also grow and that other new tools will need to be developed over the next one or two quarters.
Many small improvements have been made to the AMP system and will continue to be required during the next months. The two most important tasks to be undertaken in the immediate future are:
Both of these are difficult problems and will require research to address them. Because the major effort over the next quarter will be the further deployment of AMP monitors progress towards these goals may be somewhat slowed but remains important.
Tony McGregor continues to lead the AMP activity. A student (Ryan Kassel) is helping in the coordination of the deployment of AMP monitors while continuing to support the AMP staff in contacting new sites; his main duties are to contact HPC sites for information requests and to keep track of the sites and their information once they have received an AMP monitor. Ryan is also currently working with Tony on the general presentation and additions/changes to the AMP web pages to make them more useful and beneficial to the HPC community and researchers in general as well as continuing to manage the deployment of new AMP sites as they request AMP monitors.
After selecting a site at http://moat.nlanr.net/Active/, a menu of latencies and losses relative to the selected site appears, such as:
Detailed graphs can be obtained by selecting individual sites:
which allows for further zooming in to individual dates:
as well as into "routes" information:
Multiple different traceroute outputs can be displayed with CAIDA's "Otter" tool:
another more interesting Otter example:
A throughput test page exists, but access to the throughput tests themselves is controlled, due the opportunity for abuse:
Last quarter a start on a proposal for a new protocol to support active measurement was made. This protocol proposal has been further refined and a `straw man' draft has been written and published via the web (at http://moat.nlanr.net/ActMon/IPMP). There have been a small number of very positive responses to the protocol indicating that there is a real need for a protocol to fill this role. The responses have include offerers of support should NLANR decide to work towards standardizing the protocol through the IETF. To do so would require a substantial input of resources and careful thought will be given to this decision over the next few months.
The simulations of HTTP over asymmetric satellite connections mentioned in the last quarterly report were well received at the SPIE conference in Boston. As a consequence of the positive results they showed another series of simulations were undertaken this quarter to investigate whether similar gains could be made on sub-oceanic international fiber connections. The study shows a nearly 20% improvement in simulated http fetch times and was published at the New Zealand ATM and Broadband Workshop. The paper, which is titled "Reducing US/NZ Page Latencies" may be found at: http://moat.nlanr.net/~tonym/networks99.ps.
Although this work is of a lower priority for MOAT the results are very promising for high bandwidth/delay product links carrying HTTP traffic. As such they may be particularly relevant to the HPC community and further simulations will be undertaken in the next quarter, time permitting. It is expected that the results will be submitted to an international conference.
Jeff Brown, developer of the Cichlid software, re-did the Cichlid web page and added some demos, available via http://moat.nlanr.net/Software/Cichlid/. He is also working on a new version of Cichlid.
Todd Hansen, a recently hired UCSD student, ported the Cichlid client to Windows 95/98/NT. The port is available at http://moat.nlanr.net/~tshansen/cichlid.html. The reasons behind a Windows port are the following: The client can now be run on the most popular desktop operating system in the world; we get access to openGL hardware accelerated video cards currently supported by Windows but no other platform, and we can truly say we have a cross-platform solution, with the server running on a number of different UNIX platforms.
For the future, Todd is hoping to work with Jeff Brown to develop a common GUI interface for Cichlid that works on all platforms. They are also working to develop a single server capable of supporting multiple clients concurrently. Todd is hoping to better refine the Windows version with better error messages and other usability improvements. Also we are looking at ways to compare the various openGL cards and platforms as they perform for cichlid. For instance we have noticed that the nurbs graphing runs faster under most UNIX platforms than under Windows because of its intensive math operations.
Neil Cotofana, another UCSD student,read up on the different forms of routing and organization in the Internet, particularly the classification of collections of routers into Autonomous Systems and the use of the BGP protocol to interconnect them. He briefly began to investigate using OpenGL to implement the graphics of a BGP visualization tool. Most recently, he has started to learn Perl, as it is a powerful tool for scanning through large, formatted text files, picking and choosing desired information. These efforts are working toward the eventual goal of creating a tool to visualize and analyze BGP statistical data that will use the BGP-dump files created by University of Oregon's Route Views project as the raw data source.
In the April-June time frame, Neil hopes to see the birth of a useful BGP data visualization tool with at least some, if not full, functionality.
The primary person working on PR and outreach is Mike Gannis. He revised and ordered more "NLANR Measurement and Analysis" and "NLANR Coral Monitors" posters, and revised the NLANR handout. He also planned and did preliminary editorial work on "NLANR Packets" Web Newsletter, to be published at http://www.nlanr.net/NLANRPackets/.
Current staffing consists of Hans-Werner Braun as the overall Principal Investigator. Tony McGregor (University of Waikato) is supporting MOAT by leading the AMP project. He is supported by Ryan Kassel, who intends to work full time this summer on AMP in a research capacity. Todd Hansen has been working on the Windows adaptation for Cichlid, but is expected to take on more system administration responsibilities, and also helps on the AMP activity. Jeff Brown is the principal Cichlid developer. Neil Cotofana is working on routing visualizations. Ryan Kassel, Todd Hansen, Jeff Brown, and Neil Cotofana are UCSD students, working part time for NLANR/MOAT.
Mike Gannis, supported by Edna Nerona, (both SDSC staff) are supporting NLANR/MOAT on PR and web interface activities.