Attendees
Anjum, Kashif, Raja, Les. Joun is travelling. Umar presented his apologies but unable to attend.
General
Arshad arrived at SLAC last night. Les met with the Rector and Arshad this morning. We went over the current status and the possibility of engaging with the University of Malaysia in Kuala Lumpur. The VC from KL visited NUST recently. There are 2 funding bodies one is directly from the gov, the other eScience is for IT related things. Les will keep Anjum updated, so that Anjum can contact UTM folks. UTM is very agile. It is close Singapore.
UNIMAS is starting to monitor other Malaysian hosts outside the regular Beacons. After talking to Arshad I am going to propose to them to invite 1 or more people from NUST to go over setting up multiple monitoring sites (i.e. how to and experiences), and what can be learnt from the measurements. This would be for a workshop in December
Deploy another node in Faisalbad University. Have a POP node, this will give us more information.
HEC Report - Anjum, Amber and Imdad
Kashif is taking this over. Better to prepare from 1-30th of month, then submit on 5th of following month. Curently Kashif is going through the process and figuring what is to be analyzed and presented.
Pakistani Hosts
LSE had some security issues. Anjum is investigating. He will also investigate the other monitors. LSE is delaying the process and not responding well.
FSBD POP has high unreachability values, which is not acceptable. This is improving. They are looking into it. Backhaul network is currently leased from PTCL however in 3-4 months they will replace it with their own network. There would be no commercial traffic on it. As a result it is expected that RTT and losses will improve drastically. So next 6 months are important for observing the network performance. Went from unrelaibility then to long RTT. Not moved yet. Kashif will look into long RTTs. Kashif looked at FSBD pingtable.pl data for Sep 4th and the RTT from FSDB to PWR as "too" stable to be true. It is working fine now but the large values seen just on Sep 4th are distotring the overall averages. Kashif and Les will look at.
44 nodes are working, 2 down (Pingerpwr, Pingersau (Sind Agricultural University)).
Non Pakistani Hosts
Three non Pakistan nodes were down. Jordon is now OK, Nepal deleted, and mail sent to Sri-Lanka.
TULIP
Raja was able to successfully obtain the coordinates for several input files using all three methods i.e. (Speed of internet, Speed of light and CBI best fit). He has access to all the codes. Currently he is going through the Matlab codes and following Mathematica tutorials; this will enable him to transfer the current code to Mathematica. Before this he was not familiar with Mathematica. When we understand the code it may make sense to use a modified outer circle. I.e. instead of distance (radius in km)= 100km/ms*min_RTT(ms), use distances(radius in km)=alpha_max*100km/mx*min_RTT(ms) where alpha_max is the maximum alpha for landmarks to target in a given region. This is on hold until we grasp on ths statistics
Raja has obtained the results using the existing codes on the SEECS Server. He will modify the code after he has successfully replicated it in Mathematica. Raja is done with the analysis for a while and will document his findings and then start on th Matlab to Mathematica port.
Statistica analysis of Alpha
Raja has plotted Alpha values against distance between nodes for Europe, Africa and N.America. The plots show that nodes further apart are more likely to have a relatively higher alpha value than nodes close to each other. Nodes further apart don't have very small Alpha values unlike nearby nodes which may have alpha values of 0.1 and below.
This might be because that nodes in close vicinity have more indirect (geographically speaking) communication path and more frequent hops. The path between nodes further apart is likely to have long continuous optical fibre segment/s. The more direct the path, the larger the alpha value.
When we are using Tulip, what we know a priori is the region the landmark is in and the min-RTT (we do not know the distance yet). Thus what we want is the alpha to yield the known distance plotted against the min RTT. Then we can choose the optimum alpha as a function of min RTT. Further if the min RTT is > some value then it is of no use and should be discarded. Obviously >=0.45sec is discardable since then it is probably a satellite link, and the RTT has little to do with actual distance between landmark and target. I am guessing that Tulip will only be of use when we have large numbers of landmarks in a region (i.e. N America, Europe, Pakistan, ...) and the landmarks are in the same region as the target (probably determined by the tiering). Thus I am guessing (need further study) that if the min-RTT is > 100ms the data is of little use.
Further if the min RTT is < than some number (e.g. 5msec) then we just say the target is centered at the landmark with that min RTT and with an uncertainty of alpha*min_ RTT(ms) km, where the alpha chosen is that we have seen for min_RTT <= 10ms.
It might help to look at subsets of the data with say 0-5ms, 10-20ms, 20-40ms, 40-60ms, 60-80ms, 80-100ms plotting alpha vs min_rtt and plotting a trendline for each subset .
Raja has created histograms for each region of the frequency of pairs (landmark to target) in each bin (0-10ms, 10-20, 20-30, 30-50, 50-100ms)
The alpha does not appear to have much of a correlation with min_RTT for N. America and Europe. We may be able to use the ranges (min and max) with the Constrained CBG. Don't bother with Africa. Raja has added Pakistan. Pakistan has a very rich set of landmarks taht make it an excellent proving ground.
Raja produced an interesting graph of Distance versus alpha. This shows a roughly linear correlation for some points. Lines of constant min_RTT show up as straight line through the origin. Thus what we are seeing is clusters of common min_RTTs. This is confirmed by looking at a frequency histogram of min_RTT.
Possible projects
Future meeting time - Les
Normal next fortnightly meeting would be on Wednesday October 10th 2012 at 8pm in US and Thursday October 11th 2012 at 8am in Pakistan.