Minutes from MB-NG Task 2 meeting. Friday 3rd May 2002 10:30 am. Present: I. Bridge S. Bhatti Y. Li A. Di Donato R. Hughes-Jones F. Saka G. Ferry Load Traffic ------------ Control logging. How do we start and stop a test. How long does it take the network to reach a stable state? Measuring load as a function of time. Both for control, background and test traffic. QUESTION:How do we get data from the core routers. Do we need PCs at POP? ISSUE: SNMP-Agent may not be able to keep up with the card (not all frames and errors are reported). Thus we may not be able to completely trust it's reports. (Saleem metioned that media servers for playback are available at UCL.) We need reproducible background. UDP for analytical TCP for demonstration. (Statistically reproducible). What is the capture time we can playback? Ian Bridge says we can capture up to 15 seconds of traffic with an analyser. QUESTION: Is this enough? QUESTION: How do we know when background has stabalised. Need to assess this by measuring the transmission through the network for various network states and decide when the steady state is reached. QUESTION: Where is the background traffic directed? Measurements should be performed "cross traffic" and "dumbell"(through traffic) Generating MPLS frames. Can it be done at end nodes? Where are MPLS labels put? Meaasurement system ------------------- - MB-NG needs o Technical and political - Task 2.2.1 define low level probe measurements - Metrics pragmatics - tools - Control logging Measurements must be performed on "Empty System". and work outwards. Understand parts. (Links and nodes) - PC-to-PC - Routers and LANs - Development core. Useful to know GSR performance details. How do we load it? - End-to-end o We want to normalise system o We want to be precise about how we present because of competing projects. o We want Unidirectional and bidirectional measurements. o Uni-directional measurements implies synchronised clocks. ISSUES: We must agree on: What tools we need. Who will work on what. Defining measurements: ---------------------- - What do we need to measure for MB-NG? - How do we analyse it? - What do we expect? - What does it tell us? General conditions. - IP - MPLS - Diffserv These require input from the policy (Task 5) - Check queuing policy for equipment. e.g WFQ - ECN scenario quentions & answers mus be specified. - UDP-lite. A new IETF draft. may be worth looking at. - non-TCP protocol Suggested measurements by Richard Hughes-Jones. - UDP round trip latency vs. Packet size. - UDP round trip histograms - UDP one-way delay - UDP throughput vs. packet size and transmit delay - UDP packet loss vs. transmit rate and burst size - UDP inter-frame jitter as function of packet and transmit spacing. - TCP round-trip latency vs. message size - TCP round-trip histograms - TCP throughput vs. message size and transmit delay - TCP throughput vs. Window size. (TCP tuning) - TCP throughput vs. no of streams (same pairs. High no of streams give bursts) - TCP protocol behaviour Required measurements suggestion by Saleem Bhatti: Make a table of measurements against traffic patterns. Each meaurement with a specified traffic pattern must be justified otherwise we do not perform it. ACTION: Richard Hughes-Jones will encorporate this method more formally. IPPM metrics ------------ - A. Di Donato talked about IPPM metrics. Measurements of one-way delay is a key measurement. It is required for other measurements including the definition of when a packet is considered to be lost as opposed to simply delayed. ISSUE: The ideal equipment seems to be GPS, however it may require an aerial to be mounted on top of the buildings. ACTION: Ian Bridge to look into GPS cards for PCs. ISSUE: Simulation? This aspect of the project has been overlooked. It requires at least one person full time. ACTION: UCL continiue work on metrics. ACTION: Manchester List of measurements wanted. ACTION: Richard Hughes-Jones will encorporate tabular format of measurements, traffic patterns and their justification. ACTION: I. Bridge Look at hardware one-way delay measurements (GPS cards). S. Bhatti will setup a tech-MBNG mailing list. Network entropy. Network temperature measurement, Website: mbng.net There is a possible clash with UCL policy which states that all web site hosted at UCL must have UCL in the url. If this is true, R.Hughes jones has offered to host it at Manchester. Outstanding questions and issues. --------------------------------- 1) QUESTION: How do we get data from the core routers. Do we need PCs at POP? 2) ISSUE: SNMP-Agent may not be able to keep up with the card (not all frames and errors are reported). Thus we may not be able to completely trust it's reports. 3) What is the capture time we can playback? Ian Bridge says we can capture up to 15 seconds of traffic with an analyser. QUESTION: Is this enough? 4) QUESTION: How do we know when background has stabalised. Need to assess this by measuring the transmission through the network for various network states and decide when the steady state is reached. 5) QUESTION: Where is the background traffic directed? 6) ISSUES: We must agree on: What tools we need. Who will work on what. 7) ISSUE: The ideal equipment seems to be GPS, however it may require an aerial to be mounted on top of the buildings. 8) ISSUE: Simulation? This aspect of the project has been overlooked. It requires at least one person full time. Next meeting: 22nd May 2002 in Manchester.