User login

Weekly Report

05

Jun

2014

Making the path builders work proved far harder than I thought possible, for just implementing a fairly simple algorithm.

I have three systems now, one which just finds any two minimally overlapping paths. This was fairly simple to put together. The second was the one that proved the most difficult, and it involves having two minimally overlapping paths between each pair of nodes using only two flows per switch. The third doesnt create full paths, instead for each node it has two paths it can use to send packets to a given destination, each leading to another node that is closer to the destination, or to the destination itself.

Yeah, implementing the second one of these turned into a bit of a nightmare, what I thought would be a one afternoon job took me all week. So many little edge cases with no elegant solutions I could find all turning the code more and more labyrinth like. It wasnt a very fun week.

Anyway, its all going now, so I am going to do comparisons based on the number of flows, the length of paths found by these algorithms, and how effectively they distribute packets around the network.

Then its mostly writing from here on. I would like to have a full network running this that I can introduce some loss into and test how quickly I can locate the problem and work around it and how many packets get lost in the process, but that seems like a distant goal at the moment considering the time I have left.