1. IEPM-BW: Bandwidth Change Detection and Traceroute Analysis and Visualization Connie Logg, Joint Techs Workshop February 4-9, 2006

2. BW Change Detection: Important Know what you are looking for How long must a change persist before alerting? What threshold to use for alerting (drop of N %)? What probes provide quality data and are relevant? May differ between network types and technologies Once an alert is detected, what circumstances must be met before another alert is generated for the same or new drop? Alerting and forecasting/predicting future performance are two different things – however data taken may be relevant to both Remember – Don’t want to respond to every little glitch – more probing may escalate a minor momentary congestion event.

3. What to do with ALERTS Study them for accuracy and relevance What information would help diagnose the drop? Were there traceroute changes? Do changes in other probes seem to have occurred in the same time frame? Was there an increase in the ping RTT times? If TCP RTT is available, was there a change in that? What does OWAMP show (to be implemented)

4. Algorithm - Simplified Stream of data t 0 - - - t n 2 buffers: history buffer (hbuff) and trigger buffer (tbuff), sizes hmax & tmax Load data t 0 -t hmax into history buffer and calculate baseline histmean(hm) & histsd(hsd)

5. Algorithm - Simplified Loop over data t = {t hmax+1 - - - t n } 1.if t > hm -2*hsd, tbuff oldest ->hbuff, t->hbuff, drop hbuff oldest,calc hm & hsd, next 2.If t tbuff If size(tbuff) < tmax, next; Calc tbuff mean (tm), if (hm-tm)/hm > threshold, generate an alert, tbuff -> hbuff, calc hm, hsd, next Once alert is generated, drop threshold must be met again from the tm or the data stream must recover for ½ of drop time.

6. Overview What we currently look for Look for a drop lasting at least 6 hours Look for a drop of 33% Before reporting another drop, require 3 hours of restored throughput  Time  Bandwidth 33% drop 6 hours 33% drop 6 more hours Up for at least 3 hours Drop of 33% for 6 hours

7. Observations: Traceroute changes occasionally coincide with bandwidth drops Challenge: How do you defined a traceroute change and which have most priority? Checksum error Duplicate responding or non responding hop ! Annotations IP addr differ in 4 th octet (or 3 rd and 4 th octets) How do you quickly review traceroute changes?

8. Traceroute Visualization One compact page per day One row per host, one column per hour One character per traceroute to indicate pathology or change (period(.) = no change) Identify unique routes with a number Inspect the route associated with a route number Provide for analysis of long term route evolutions Route # at start of day, gives idea of route stability Multiple route changes (due to GEANT), later restored to original route Period (.) means no change

9. Pathology Encodings Stutter Probe type End host not pingable ICMP checksum Change in only 4 th octet Hop does not respond No change Multihomed ! Annotation (!X) Change but same AS

10. Navigation traceroute to CCSVSN04.IN2P3.FR (134.158.104.199), 30 hops max, 38 byte packets 1 rtr-gsr-test (134.79.243.1) 0.102 ms … 13 in2p3-lyon.cssi.renater.fr (193.51.181.6) 154.063 ms !X #rt# firstseen lastseen route 0 1086844945 1089705757...,192.68.191.83,137.164.23.41,137.164.22.37,...,131.215.xxx.xxx 1 1087467754 1089702792...,192.68.191.83,171.64.1.132,137,...,131.215.xxx.xxx 2 1087472550 1087473162...,192.68.191.83,137.164.23.41,137.164.22.37,...,131.215.xxx.xxx 3 1087529551 1087954977...,192.68.191.83,137.164.23.41,137.164.22.37,...,131.215.xxx.xxx 4 1087875771 1087955566...,192.68.191.83,137.164.23.41,137.164.22.37,...,(n/a),131.215.xxx.xxx 5 1087957378 1087957378...,192.68.191.83,137.164.23.41,137.164.22.37,...,131.215.xxx.xxx 6 1088221368 1088221368...,192.68.191.146,134.55.209.1,134.55.209.6,...,131.215.xxx.xxx 7 1089217384 1089615761...,192.68.191.83,137.164.23.41,(n/a),...,131.215.xxx.xxx 8 1089294790 1089432163...,192.68.191.83,137.164.23.41,137.164.22.37,(n/a),...,131.215.xxx.xxx

11. AS’ information

12. Changes in network topology (BGP) can result in dramatic changes in performance Snapshot of traceroute summary table Samples of traceroute trees generated from the table ABwE measurement one/minute for 24 hours Thurs Oct 9 9:00am to Fri Oct 10 9:01am Drop in performance (From original path: SLAC-CENIC-Caltech to SLAC-Esnet-LosNettos (100Mbps) -Caltech ) Back to original path Changes detected by IEPM-Iperf and AbWE Esnet-LosNettos segment in the path (100 Mbits/s) Hour Remote host Dynamic BW capacity (DBC) Cross-traffic (XT) Available BW = (DBC-XT) Mbits/s Notes: 1. Caltech misrouted via Los-Nettos 100Mbps commercial net 14:00-17:00 2. ESnet/GEANT working on routes from 2:00 to 14:00 3. A previous occurrence went un-noticed for 2 months 4. Next step is to auto detect and notify Los-Nettos (100Mbps)

13. New Graphical Map Display New Traceroute Map Display

14. Quality Control – Bandwidth Monitoring It is good to have a local target host for a sanity check: Problem here was that the monitoring host rebooted into single CPU mode after maintenance had been performed on it.

15. More Sanity Checks Target host – iepm-bw@caltech – was not completely installed – process cleanup did not have the perl modules that it needed to kill lingering processes (needs install check)

16. Probe Correlation Pathchirp analysis shows drop Multi- stream iperf show drop Single stream iperf shows drop Traceroute change affected all 3

17. Analysis Results Email is sent to interested parties with links to graphs, data and traceroute analysis Alerts are saved in the ALERT table and graphs are saved in the GRAPH Table for future reference Every analysis run, about every 2 hours, a table showing which alerts occurred for which probes and when is generated. It has links to the more detailed alert information. Reports are generated nightly for the last month alerts from these tables.

18. Future Improvements Integrate Ping RTTmin and RTTmax analysis Optimize code for speed of execution – estimate mean and std dev Upload alerts to MonALISA – What info? Compare detection algorithms (KS, HW, PCA?) Recommendations on data taking frequencies and how to define the trigger and history buffer sizes still needs more exploring Implement prediction/forecasting algorithm(s) QUESTIONS?