Meeting Notes 1/11/2008
* CTP problems * Testing DIP
* Phil, Stanford * Razvan, JHU * Kaisen, UCSD * Om, USC
* Kaisen: check in tests/DIP and send an email * Om: Do more investigation on CTP problems
(This is a continuation of our discussion on the mailing lists last week)
Om: Lets look at the results from the 24-hr CTP run. The problem is starting around 30k seconds.
Om: Lets look at the subdirectory that has detailed results from this run.
Phil: node 29 has the most wait acks, node 24 drops all the pkts. This must be a memory leak somewhere.
Om: Maybe, lets run the find-leak.pl script.
Phil: Does not seem like there is a leak. But starting unix time 1199337795.769598, node 24 does a lot of wait ACk's but is not able to get the pkt to its parent. The parent happens to be node 0. So, it is hammering on this link with no success.
Om: Maybe there is some corner cases related to node 0 that is causing node 24 to behave this way. I will look at this offline and send emails.
Kaisen: I have test setup in which node 0 sends new data, each node knows a pre-determined number of messages, and upon receiving the expected number of updates, it sends a message - so if we receive that serial message from all the nodes, we know that all the updates were disseminated across the network.
Om: DIP is designed to disseminate a large number of data items. Can I test it with 100 data items?
Kaisen: No. But you can specify the number of data items and don't want to do too many data items because it will consume a lot of memory.
Phil: It does not matter if it consumes a lot of memory because this is just testing and while testing, we want to test as many items as possible.
Om: will check in test program to tests/TestDIP with appropriate readme and send email out
Phil: It will be great if there is some LED status indicators while the test is going on so that the user knows if it the test was successful, etc.
Kaisen: Don't want to do LEDs because of energy consumption
Phil: For testing DIP, energy consumption does not matter.