The development and ongoing management of the network analysis infrastructure (NAI) continued with excellent progress this quarter. Additions and upgrades were made to both the passive measurement infrastructure and the active measurement mesh. Several significant measures were taken to strengthen and improve the efficiency of the AMP system (mesh and data collectors). Major progress was made with the IPMP protocol implementation.
Our outreach activities, including collaborations and papers, were highlights this quarter. Several new collaborations have been, or are being, developed, including working with researchers and administrators at Internet2 (both the Active Measurement Project (AMP) and the Passive Measurement and Analysis (PMA) project are planning collaborative activities). The NLANR Traces User Community continues to provide useful feedback and discussion. New graphs were presented to them which satisfied several requests; a number of other requests from the community for data collection with different parameters are under consideration.
Several well received papers were presented at conferences and workshops this quarter, including three at the Passive and Active Measurements meeting (PAM2001) in April. Six papers were submitted to the ACM SIGCOMM Internet Measurement Workshop (IMW2001) to be held in San Francisco (November 2001). Of particular note, five students were closely involved (three as first authors).
Students continue to make significant contributions to the project in many areas (script writing, displaying data graphically, Web page development and maintenance, technical support regarding data, as well as other activities).
Considerable thought and numerous discussions regarding future directions and strategies for the Network Analysis Infrastructure (NAI) were conducted. Strategies and plans for new directions in the PMA and AMP projects were identified and developed.
The passive monitoring infrastructure continues to be developed. A passive monitor was deployed this quarter (at the University of Buffalo); additional monitors are in preparation for installation and deployment. The use of the High Performance Storage System (HPSS) at SDSC for the storage of collected PMA data has now been fully implemented, and is an ongoing process.
Upgrades of the optical carrier connections at various sites are beginning. This is expected to create the need for PMA monitors with Dag3 interface cards (for sites deployed before the new Dag3 technology became the core of the passive monitors).
Progress on the router instrumentation has been hampered by the availability of OC48c measurement gear. Very preliminary work began regarding sampling; initial sampling research was done using one PMA traffic trace. The long term plan is to iterate this process using a number of PMA traces.
PMA has made a number of important contacts for collaboration on the analysis of measurement data and the installation of new sites, specifically for international access links. Further PMA activities are detailed in the documentation/tools, papers, and collaborations sections, following. Please see the Web pages for more information on the PMA project: http://moat.nlanr.net/PMA/.
A major project to reimplement some of the PMA tools from Perl to C began this quarter. A revamp of the PMA data collection procedures has become crucial due to the outage of our main trace collection server. Reimplementation of several of the Perl scripts into C must be done very soon. Figure 1 below shows the performance comparison between the Perl and C implementations of tsh2ta.
Figure 1. This performance comparison graph of the Perl and C implementations of tsh2ta shows a dramatic improvement with the C implementation. For more information, please see http://moat.nlanr.net/~dgcheney/tsh2ta/Stats.html.
The data collection functionality of AMP continues to mature and maintain a high level of reliability. However, sites in the AMP mesh are entering a phase of change. Changes resulting from sites switching between HPC providers (to I2/Abilene systems from vBNS, for example) have caused some anomalies to occur. Occasionally, routers start advertising commodity Internet paths rather than the intended I2/Abilene path. These anomalies have been easy to correct, but will require some attention as these changes progress. Use of the HPSS at SDSC for AMP data collection has fully matured and is an ongoing function.
Decisions were made on the new hardware options to be used in the AMP monitors. Therefore, additional monitors were deployed to remote sites (including two sites outside the US: Canada [Canarie] and Korea [Kordic]). This brings the total number of AMP monitors to 121. The new machines are 1U in size and very compact. They have an on-board ethernet adapter, which is not supported by our operating system. An Intel ethernet adapter is being used on a temporary basis until we upgrade to a newer version of the operating system. We hope to ship machines using the on-board ethernet (resulting in a machine that has no expansion cards).
The system manager tool is developing well. There has been further progress on driving updates directly from the database. By deploying it over a larger number of machines, we are beginning to understand the full impact of the system manager with respect to bringing new machines up to speed. To complete the software set-up of newly deployed monitors, we now have just a short list of steps to perform, then the system manager handles the other (software) tasks required for system set-up.
Several significant measures were taken to strengthen and improve the efficiency of the AMP system (mesh and data collectors), as follows:
Testing with regard to the AMP monitoring function with the FreeBSD 4.2 OS version proceeded this quarter. The tests did not confirm an improvement in the network interface throughput; however, a number of other advantages to the upgrade were identified. It is expected that a decision to upgrade to FreeBSD 4.2 is forthcoming.
Major progress was made with the IPMP protocol implementation. The kernel code has matured and is stable; the ipmp_ping client is also mature and stable. The information request/reply facility of IPMP is under development. Experiments have been conducted using NTP as an external clock source for taking one-way delay measurements with known accuracy limitations. A paper on IPMP describing the limitations of existing active measurement protocols, the IPMP protocol, and how IPMP addresses these limitations was written and submitted to ACM SIGCOMM.
Figure 2. This graph shows the delay in round trip time (RTT) of ICMP relative to the RTT of IPMP. As can be seen, delays (fairly significant) begin around noon and continue to midnight.
A massive amount of work was done during the beginning of AMP, including the design of the data collectors and the actual placement of echo hosts at many HPC universities. The data collector software has worked well to the present. With the development and implementation of IPMP in the AMP infrastructure, we are able to do more sophisticated measurements which retrieve additional performance information from the network with a relatively simple packet exchange. Packet delay, clock adjustments on echo hosts, packet loss, and route asymmetry in terms of both hop count and potential delay are some of these parameters.
This ability to glean additional information from the network with IPMP required a redevelopment of the data collector infrastructure to enable the collection of this information in a scalable manner. The measurement daemon is designed to handle the collection of this information in a single executable. It schedules the measurements, collects the data, and performs compression on the data collected so that the data is stored in an efficient manner.
In addition, we wish to share our infrastructure with other measurement groups in the future. In order to allow this to occur in a manner that does not create a significant support overhead, a single executable with a well designed configuration interface will allow us to do this in an elegant fashion.
A measurement daemon designed for measuring meshes with IPMP is being developed. It will replace all of the various Perl, cron, and ping clients which were quickly done in the early stages of AMP. Current tasks include defining and implementing a file format that allows us to gain compression by identifying redundant information in the file.
Some progress has been made with the on-demand and regular iperf tests over the AMP network. Results have been slow due to a number of factors, including unexpected issues with making the scripts robust to errors. It is expected that on demand tests and regular iperf tests will be running within the month. iperf is a heavy weight throughput test, produced by NLANR DAST; it is being used to provide some insight into throughput results and how they are dependent on round trip time (RTT) results, and other parameters. These activities are in support of the research into development of a light weight bandwidth estimator, which is the ultimate goal.
Additional AMP activities are detailed in the documentation/tools, papers, and collaborations sections, following. Please see the Web pages for more information on AMP: http://moat.nlanr.net/AMP.
- Wireless Network
Performance measurement and analysis activities continued on the wireless network (HPWREN). These involve MIB data from the Lucent and Tsunami Radios, and the extraction of the SNMP data from them. The MIBs show vital information such as link health, link usage, signal and noise levels. Factors affecting network performance can be revealed using this data; this data can also be used to correlate with the corresponding weather data.
Installation and field work was performed to redeploy the weather stations (at both sites). Redeployment was necessary because the weather station design was improved (reconfigured for better results). Numerous throughput, round trip time (RTT), and received signal level (RSL) measurements were performed in the field (in a variety of settings and conditions). These tests were conducted using the Lucent COR and ROR in different modes (COR, ROR Master-slave, IEEE 802.11 Access Point mode), with and without encryption.
New network analysis scripts for the MIB data were written (and a few existing ones were modified), and incorporated into the Web pages (http://stat.hpwren.ucsd.edu/). Two papers were submitted to ACM SIGCOMM (November 2001) regarding the performance measurements on HPWREN (please see the papers section of this report for more detail).
- Network routing
Network routing activities for this quarter were focused on the collaborative effort with the SKIDL group (the evolving data mining group at SDSC). The purpose was to explore the application of data mining techniques and commercial tools to networking data. Specifically, GateD BGP routing data (currently collected and archived by NLANR) was analyzed. Considerable time was devoted to characterizing the data and deciding what type of analysis could be performed, given the nature of the data set. Ultimately, the collaboration used Perl to format the data, SQL databases, and Intelligent Miner (IBM's commercial data mining software) to generate a model for specific phenomena evident within the data, and then tested the accuracy of that model.
Documentation and tool development for all aspects of the Network Analysis Infrastructure (NAI) is ongoing.
The initial prototype site summary graph for AMP, developed last quarter, has been deployed. This graph allows users to see changes in groups of machines over time. All sites will be graphed based on their average measurements each week. As this matures, it will be possible to identify subsets of sites that are affected by a change, or event. See Figure 3 below.
Figure 3. Prototype of the new route display graphic. larger view
A major project for reimplementation of several of the Perl scripts into C has begun. The first tool to be reworked was the tsh2ta script. For additional detail, see the PMA activities section and Figure 1 above for a performance comparison of the implementations.
Technical support regarding our PMA formats, methods, and practices in response to questions submitted to us by researchers using the passive data continues. This activity will aid in further development of the Data Format FAQ for the PMA project, as well as providing a human resource to answer questions (especially for researchers looking for data with very particular conditions and parameters).
The Cichlid 3-D visualization system continues to be developed. Work on compiling Cichlid for Windows, and giving it a GUI front-end was performed this quarter. The Cichlid for Windows and the Cichlid server repository Web pages were updated. And a working bar chart version of the new Cichlid server that displays use of bandwidth on HPWREN has been developed; work continues on the vertex/edge iteration.
A new issue (print/pdf version) of the Network Analysis Times was published; it can be viewed at http://moat.nlanr.net/NATimes/april2001.pdf. It was distributed at many conferences and meetings held during the quarter. In addition to the those listed in the following meeting participation section, the NATimes was distributed at:
At most of the conferences and meetings, the NLANR Packets summary pages and various posters (11x14 posters of AMP, PMA, and NLANR) were also distributed. A new NLANR Packets is in development. Please see the Web page: http://www.nlanr.net/NLANRPackets/.
Preliminary plans are being developed for NLANR Measurement and Analysis Group demos for the SC 2001 conference. A Web page for the HPIIS workshop was also created - http://moat.nlanr.net/Workshops/HPIIS-2001/.
Several papers were written and submitted this quarter, including six papers (representing nine different members of the group) submitted to the ACM SIGCOMM IMW2001, Internet Measurement Workshop (November 2001, San Francisco). Papers presented this quarter are listed in the conferences/meetings section following.
An extended version of the following paper was included in the publication of the extended versions of the best papers presented at SPECTS'99.
The NLANR Measurement and Network Analysis Group was represented this quarter at the following meetings:
Passive and Active Measurements (PAM2001) - Amsterdam, the Netherlands, April 23-24, 2001
STAR TAP International Advisory Committee Meeting, INET 2001 - Stockholm, Sweden, Tuesday, June 5, 2001
NRDM workshop - Santa Barbara, CA, May 2001
14th ITC Specialists Seminar on Access Networks and Systems - Barcelona/Gerona, Catalonia, Spain, April, 25-27, 2001
NLANR/Internet2 Techs Workshop - 14-17 May 2001, Lincoln, Nebraska
New Zealand Computer Science Research Students' Conference, University of Canterbury, New Zealand, 19-20th April 2001
We began plans for hosting a workshop in August at the San Diego Supercomputer Center (SDSC) in conjunction with NSF's High Performance International Internet Services (HPIIS) Project. The workshop will focus on developing measurement metrics to quantify the use of HPIIS links in addition to possibly identifying research collaborations. Domestic and international advanced networks will be involved. In addition, we are planning to host a small demonstration on how to create your own AMP network. (The Canarie network [Canada] and the KORDIC Supercomputing Infra Research Lab [Korea] have both expressed interest in creating their own AMP meshes.) For more information, please see the Web page: http://moat.nlanr.net/Workshops/HPIIS-2001/.
We also maintain numerous continuing collaborations; the following list includes both long-term relationships, as well as recently developed ones. This quarter we continued work with:
Pavana Yalamanchili, UCSD graduate student, completed her Masters's thesis on "Study of Unlicensed Band Wireless Data Network," defended, and graduated this quarter. She continues to work with us.
Jamie Curtis (University of Waikato graduate student) presented his paper entitled "Review of Bandwidth Estimation Techniques" at the New Zealand Research Students Conference (end of April). It placed in the top six (6) papers and was therefore accepted to a special edition of the New Zealand Journal of Computing, now published. See papers and presentations section for detail.
In preparation for beginning graduate school (PH.D. student in geophysics) at the Scripps Institute of Oceanography (UCSD) this Fall, Jose Otero worked with Frank Vernon and Cathy Constable on a number of GPS related activities. This activity will eventually lead to more efficient means to align earthquake sensors. This included work on the Quanterra data logger.
David Cheney, UCSD undergraduate student, continues to work in a technical support capacity for the PMA project: answering questions submitted to us by researchers regarding our formats, methods, and practices. This activity helps to inform his work on the Data Format FAQ for PMA.
Of the six papers submitted by the group to the ACM SIGCOMM IMW2001 workshop, five students were closely involved (three as first authors).
Passive Measurement and Analysis (PMA)activities:
Active Measurement Project (AMP) activities:
Measurement and Analysis on the wireless network (HPWREN):
-30-