midd:traffic_scheduling:metrics
Network Measurement Metrics
These are my current notes on network metrics. These will be reformatted soon.
From class and our brains: jitter heartbeat DNS DHCP ping simulate different traffic types, e.g. streaming multimedia frequency of dropped packets amount of unacked data network saturation (achievable bandwidth) ---------- Google search ---------- Network Metric Report http://www.geant2.net/upload/pdf/GN2-05-265v4-Deliverable_DJ1-2-3_Network_Metric_Report.pdf Measurement dimensions: Performance: * Availability * Loss and errors * Delay * Bandwidth Miscellaneous: * Device-specific * Flow * Routing SNMP measure bit errors with communicating synchronized PRNGs measure bit errors at physical and data link layers one-way delay, jitter, round trip time bandwidth: capacity, utilization, available bandwidth, achievable bandwidth capacity: use snmp (hardware endpoints should know capacity) measure delay with synchronized clocks; problematic ---------- Non-metric coordinates for predicting network proximity http://research.microsoft.com/pubs/73318/coords_infocom2008.pdf Doesn't look very helpful as I'm not sure how min-plus applies to our situation. Also, overall it is an optimization procedure (iterative over time) which really doesn't help us. ---------- Loss Network Models and Multiple Metric Performance Sensitivity Analysis for Mobile Wireless Multi-hop Networks http://www.isr.umd.edu/~vahidt/LossNetworkModel_WICON08.pdf TODO ---------- Inference and Labeling of Metric-Induced Network Topologies http://www.cs.bu.edu/fac/byers/pubs/tpds.pdf hop-count bottleneck bandwidth packet loss rate estimation of shared loss TODO ---------- Controlling Multimedia QoS in the Future Home Network Using the PSQA Metric http://comjnl.oxfordjournals.org/cgi/reprint/49/2/137 TODO ---------- Improving network anomaly detection effectiveness via an integrated multi-metric-multi-link PCA-based approach http://www3.interscience.wiley.com/cgi-bin/fulltext/121481651/PDFSTART TODO ---------- IMC '08 ---------- A measurement study of a commercial-grade urban wifi mesh http://portal.acm.org/citation.cfm?id=1452520.1452534&coll=portal&dl=ACM&type=series&idx=SERIES10693&part=series&WantType=Proceedings&title=IMC&CFID=26429339&CFTOKEN=74708467 SNMP number failed transmissions, wireless noise floor iperf (TODO) ---------- IMC '07 ---------- Characterizing residential broadband networks http://portal.acm.org/citation.cfm?id=1298306.1298313&coll=portal&dl=ACM&type=series&idx=SERIES10693&part=series&WantType=Proceedings&title=IMC&CFID=26429339&CFTOKEN=74708467 queue lengths, seems residential ISPs have large traffic queues which delay packets by milliseconds do we account for upstream vs. downstream differences? 1488 byte TCP ACK packets floods, trickles ---------- Studying wireless routing link metric dynamics http://portal.acm.org/citation.cfm?id=1298306.1298352&coll=portal&dl=ACM&type=series&idx=SERIES10693&part=series&WantType=Proceedings&title=IMC&CFID=26429339&CFTOKEN=74708467 expected transmission count: how many transmissions do you expect to need to successfully transmit a packet 1/(P(packet success)*P(ack success)) measure bandwidth via packet pair, back-to-back probe packets of increasing size ---------- IMC '04 ---------- Self-configuring network traffic generation http://portal.acm.org/citation.cfm?id=1028788.1028798&coll=portal&dl=ACM&type=series&idx=SERIES10693&part=series&WantType=Proceedings&title=IMC&CFID=26429339&CFTOKEN=74708467 Related work, nothing helpful on metrics ---------- Strategies for sound internet measurement http://portal.acm.org/citation.cfm?id=1028788.1028824&coll=portal&dl=ACM&type=series&idx=SERIES10693&part=series&WantType=Proceedings&title=IMC&CFID=26429339&CFTOKEN=74708467 precision: what data is kept/omitted; timing; justify your measurement decisions associate meta-data with measurements accuracy: e.g. what can go wrong with packet filtering, several levels can drop; stable clock rate misconception: are you measuring what you think you are measuring?; beware of proxies; large socket buffers and small transfer sizes; vantage point; representativeness (i.e. controlled experiments) compute connection size by difference in TCP SYN/FIN/RST sequence numbers calibration: outliers, spikes often suggest measurement problems; self-consistency checks; multiple ways of measuring same thing; test using synthetic data ---------- Ten fallacies and pitfalls on end-to-end available bandwidth estimation http://portal.acm.org/citation.cfm?id=1028788.1028825&coll=portal&dl=ACM&type=series&idx=SERIES10693&part=series&WantType=Proceedings&title=IMC&CFID=26429339&CFTOKEN=74708467 available bandwidth metric is link capacity above and beyond average utilization mind the relation between probing stream duration and averaging time scale packet trains are better than packet pairs beware cross-traffic burstiness beware multiple bottlenecks beware iterative probing does not necessarily converge to a single number available bandwidth should not be compared to bulk TCP throughput ---------- Single-hop probing asymptotics in available bandwidth estimation: sample-path analysis http://portal.acm.org/citation.cfm?id=1028788.1028831&coll=portal&dl=ACM&type=series&idx=SERIES10693&part=series&WantType=Proceedings&title=IMC&CFID=26429339&CFTOKEN=74708467 explains how to analyze packet trains ---------- Bandwidth estimation in broadband access networks http://portal.acm.org/citation.cfm?id=1028788.1028832&coll=portal&dl=ACM&type=series&idx=SERIES10693&part=series&WantType=Proceedings&title=IMC&CFID=26429339&CFTOKEN=74708467 nothing significant ---------- SIGCOMM '91 ---------- A control-theoretic approach to flow control http://portal.acm.org/citation.cfm?id=115992.115995&coll=portal&dl=ACM&type=series&idx=SERIES419&part=series&WantType=Proceedings&title=COMM&CFID=26429339&CFTOKEN=74708467 packet-pair technique: measure ack delay to estimate server's processing rate
SNMP
Here is a list of SNMP variables to query for additional information on network connection characteristics.
- RFC 1213; Management Information Base for Network management of TCP/IP-based internets: MIB-II
- system
- interfaces
- ifTable contains entries for each network interface
- ifEntry fields including ifSpeed, ifInErrors, ifOutErrors, ifInDiscards, ifOutDiscards
- at (address translation)
- ip
- ipInHdrErrors
- ipInAddrErrors
- ipInDiscards
- ipInDelivers
- ipOutDiscards
- ipReasmFails
- icmp
- tcp
- tcpInErrs (?)
- udp
- updInErrs
- egp
- transmission
- snmp
- RFC 1514, 2790; Host Resources MIB
- hrMemorySize
- hrProcessorLoad
- More…
References:
3/31/2009
PK - it looks like SNMP won't be usable by us. a) It is hardware dependent - you must know what hardware you're talking to in order to use it b) it requires a known “community name” which works as a password - since we don't know it, we can't use it. :(
midd/traffic_scheduling/metrics.txt · Last modified: 2009/03/31 14:41 by perryk
