PingER archive site -- FYP (Ghulam and Farhan)

  • This is basically complete but yet not deployed due to possible change of schema. Once we finalize the schema, it wont take long to copy data to the database. A meeting will be setup among Dr. Anjum, Ghulam, Farhan and Bilal from SEECS and Dr. Les, Faisal and Zafar from SLAC. (Done, attended by Zafar, Faisal from SLAC and Bilal, Farhan from SEECS).
  • Future and deployment at SLAC (will be handled by Sadia and Zafar).
  • Task: Our new PingER schema will be brought closer to PerfSONAR PingER schema. This will introduce similarity helping us develop modules to retrieve information from both the projects. This schema will be profiled against our current PingER schema and flat file architecture. Ghulam and Farhan please include disk usage and processing time for 50, 100, 1000 and 10000 files.
    • Note: Ghulam and Farhan will start on this asap. They have to submit their project documentation and present final defense. This might take two weeks. Both have been invited to join SEECS as RAs for 3 months at-least. Dr. Anjum has the details. They will work on future projects as well: e.g. incorporating MOS and Alpha in table.pl, traceroute archive site, etc.

CBG TULIP Integration -- FYP (Bilal)

  • It is working .. CBG results aren't impressive. See test/development deployment at <http://203.99.52.38/cgi-bin/tulip-viz.cgi?target=www.seecs.edu.pk> and <http://203.99.52.38/cgi-bin/tulip-viz.cgi?target=slac.stanford.edu>
    • Accuracy improvement: use bestline instead of SOI. But bestline method requires pairs of landmarks and RTTs among them. For this we need to have sets of data among multiple landmarks. This can be kept in a database (TULIP) and can be refreshed on priority (daily, weekly). The only change in CBG would be to pick this data and perform calculations for generating bestline results. We won't need to compromise on speed as well (e.g. their won't be extra runtime queries).
  • Other issues include: doesn't work on MSIE (solved by Bilal)
  • Very slow (we should consider to *not* ping all landmarks at runtime).
    • Slowness due to 3 reasons:
      1. PlanetLab nodes not working due to cookies issue.
      2. PerfSONAR has security certificates which makes it slow to ping those nodes.
      3. There is no tiering of landmarks (https://confluence.slac.stanford.edu/display/IEPM/TULIP+Landmark+Tiering+Analysis).
    • Slowness improvement: Basically it will help selecting only those landmarks that are near to the target, thus cutting down on time. One option could be to have one landmark in the middle of Europe, America etc. to know what region the target is in and then select landmarks at runtime on that basis.
  • Bilal and Faisal will work together on this project.

Bestline approach of CBG

The problem is that TULIP provides only that data at run time by which only two constraints of CBG can run. For third constraint i.e. bestline constraint additional data is required. This additional data consists of lat,lons of multiple landmarks and their RTTs to other landmarks.

For this purpose we require a database which contains PingER data of different landmarks around the world. We have two options for collecting this data:

  1. Fetch this data from database and put in files in proper input format of bestline technique. So that CBG read input files process it and put results in files.
  2. At run time we fetch data from database and use it for processing. No need for creating multiple input files. Only one output file is created.

For this purpose we require a script for collecting landmarks information to update database entries.

  • No labels