User login

Stephen Eichler's blog

21

May

2013

Further negotiations with Caida have been carried out to allow the running of the traces with maximum packets per trace at a higher rate, 300 pps.

The Internet simulator was modified and is running. Some effort has been put into finding out how to make small runs quicker by using more cores.

The files for the Caida nodes have been prepared including the different files necessary for analysing with a maximum number of probes or the reduced limit. Two nodes with the former will be used and twentythree with the latter.

14

May

2013

A modified analysis was created for the Internet Simulator. This used existing available factor levels in a modified pattern. This was done to confirm the function of the simulator. It was noticed that the simulator generates some error messages in its native state and it would be useful to find out if these are important or what can be done them. The simulation is still running, and I note that with the smaller number of simulations that only one core on Wraith is being used.

The set up procedure with Caida is progressing and we should be probing soon.

07

May

2013

Negotiations were carried out with CAIDA arc regarding the use of some of their nodes for scamper data collection. Further optimisations of scamper for this environment have been carried out on Yoyo and the set of messages that scamper reports to the error file has been further reduced. This means that it may not be necessary to use screens to manually control each node in order to access error messages.

Scamper has also been upgraded to use a wide range of addresses when running the per destination modes.

An initial document regarding the use of the Internet Simulator has been written. I have been studying the code to determine how the initial changes may be made. A set of parameter settings for the first new scenario has been produced and some optimal values have been obtained from published information on the topic.

30

Apr

2013

Some further optimisation of scamper memory usage was carried out using test runs on Yoyo.

Negotiations with CAIDA about the use of Ark nodes has been carried out, and testing is to begin soon.

A test run of ICMP data collection is being carried out on planetlab. ICMP data collection will be carried out there, while UDP and TCP data will be collected on Ark.

Investigation into the Internet Simulator has been carried out to see why a subset of possible simulations based on the available paramenter levels was carried out. I have been learning about the diagnostic programs that come with IS0 including how to make graphs.

23

Apr

2013

Completed an initial Internet Simulator run, and began to read up on how to produce graphs from the output data.

Ran daily scamper runs on Yoyo to optimise virtual memory and probes per second usage by modifying the structure of the experiment. I ran UDP and TCP souce port MDA without ICMP echo mode. Hopefully it won't be necessary to run UDP and TCP separately. Limited windows to 60 or unlimited. Limited PPS to 300 or 200. Limited total probes to 15000 or 65000. The aim was to get about 5000 addresses processed per day so that the same can be done on the CAIDA Ar

16

Apr

2013

Procedures to determine the low throughput of planetlab have been carried out.

The diagnostic effectiveness on planetlab is also limited as screen cannot run and commands run from cron cause errors related to the requirement for a tty even though there is no password required for sudo on the nodes.

Scamper and the churn driver have both been modified to produce error logs that are not too long.

The internet simulator is being run in its native state from Tony to get used to what it does. It has been running for about five days so far, on Wraith.

19

Mar

2013

When a churn analysis was carried out it was noted that TCP and UDP analyses had failed. Subsequent to this the UDP analysis was fixed using a planetlab switch, however this did not work for TCP. To fix this a traceroute program that works on planetlab was analysed and similar socket structure was applied to scamper. This is now in the process of testing.

12

Mar

2013

The analysis method to quantify churn has been upgraded to identify groups of successors as load balancer ID rather than divergence nodes. This is because the divergence IP interface addresses can have aliases. There can also be more than one set of successors on one router.

Another scamper run on planetlab was kicked off, however I have found a snag i.e. UDP and TCP packets are not being sent out from planetlab nodes, only ICMP.

05

Mar

2013

The first true data collection cycle of planetlab has been completed and the data downloaded. The second cycle has been initiated and monitored. One of the nodes was wiped by the system before I could download the data it had collected. This change also manifested as the node having a different host key ID, which triggered a security warning from ssh.

Progress has been made in the area of dealiasing load balancers by analysing successor sets rather than load balancer IP address. In the first instance I plan to generate a new count of LBs and compare this with what I already have. Implementing this change to the collection of turnover statistics will involve a significant amount of programming.

I am hoping to get a little bit of guidance from Tony on how network data is presented to the internet simulator.

26

Feb

2013

The address data files for the planet lab runs were repaired and regenerated.

Cron scripts were written to automatically control and restart scamper data collection and a new first run was initiated. In addition more nodes were set up on planetlab as a safety margin, as various problems occur with nodes beyond my control, and this will ensure that data from 15 nodes is available.

Extra time for the planetlab slice has been requested and is now almost enough for the two month run time.

I have had an initial look at Tony's Internet Simulator 0. The code compiles OK on voodoo, however network data needs to be set up for an analysis to work. Other modes of analysis will need to be designed once I get a better understanding of how it works.