Quick Link

11/11
11/04
10/28
10/21
7/29
6/02
5/26
5/12


Progress

11/11

Weekly Goal

1. Controller hanging issue.

2. Complete the Algorithm integration.


Achieve

1. After examing the code, this issue is raised of the socket message from server, the buffer is a array of charater, if the first coming message is "route 0 19", and the following "route 1 1", the buffer will be "route 1 1'9'", '9' is the old dirty data, so the solution to this is to get the length of each message then cut the proper size of string in buffer to get rid of the dirty data.

2. Working (11/12).


11/04

Weekly Goal

1. Problem : My Custom Topology didn't work in Mininet

2. Continuous integrate the AI algorithm with Ryu controller


Achieve

1. Found that if there exist a loop in a topology, the host can't ping any others, maybe is the 'Switch-loop' issue, solution of conventional network is to apply the spanning tree, but I thought that the decision of AI algorithm automatically solve this issue.

2. Done implementing the socket and done combined controller with AI algorithm, rest work is the rule-installing part.


10/28

Weekly Goal

1. Implement the AI routing algorithm (cont' last week)

2. How to integrat with Ryu controller ?


Achieve

1. Done. Note Link

2. As the note above, preparing to implement a socket program to communicate between SDN controller and the routing algorithm, Protocol Buffer which defined a data type (as JSON, XML) for communication by google will be used.


10/21

Weekly Goal

1. Implement the AI routing algorithm


Achieve

1. Implement the genetic algorithm by referencing [11]

2. Almost done, needed to tune faster.


7/29

Weekly Goal

1. When algorithm adjust the network, does the changing interval cause problem ?

2. What's "best" path to choose ?


Achieve

1. The subset of the best choise is also better than others, the only problem is that packet did not arrive in a sequencial order, but it's the algorithm factor, it should be avoid when designing learning-algorithm.

2. The shortest is not obviously the best one, also need to consider load-balance, successful rate, packet duplicate and control packet overhead.


6/02

Weekly Goal

1. Complete learning-algorithm

2. Deploy to SDN controller


Achieve

1. The obstacle of implementation is to transform the energy function to mathmatical form, need more effort to handle the math detail, now the generic learning-algorithm instead.

2. Initially shoud collect the network informations which learning-algorithm needed, it may cause lots of time, but it's ok when start up.


5/26

Weekly Goal

1. Hopfield neural network implementation

2. What if the path set in phase I are not disjoint ?

3. How to implement in SDN Controller ? initial plane ? tune up optimization ?

4. Build a simulation with certain srt des pair and measure performance with traditional network, built-in SDN routing

Achieve

The simple implementation of HNN is create a single layer neural network, the next step is to specific the energy function which map the problem domain into a hyper-plane, which means it is specific to optimal only some kinds of problem.

hnn_naive_implementation


5/12

Weekly Goal

1. Research neural network routing Algorithms

2. How to implement in SDN controller



Achieve

1. Project site created.

2. By researching paper [10], Hopfield type neural network is include two phases:

In phase I, a set of alternative route are determined for each source-destination pair.

Phase_I_ex

In phase II, the traffic flow between each source-destination pair is optimally distributed to its corresponding alternative route.


Reference

[1] The Road to SDN, Nick Feamster, Jennifer Rexford, Ellen Zegura, 2013.

[2] Ethane: Taking Control of the Enterprise, Martìn Casado, Michael J. Freedman, Justin Pettit, Jianying Luo, Nick McKeown, Scott Shenker, Sigcomm 2007.

[3] Design and Implementation of a Routing Control Platform, Matthew Caesar, Donald Caldwell, Nick Feamster, Jennifer Rexford, Aman Shaikh, Jacobus van der Merwe, NSDI 2005.

[4] OpenFlow: Enabling Innovation in Campus Networks, Nick McKeown, Tom Anderson, Hari Balakrishnan, Guru Parulkar, Larry Peterson, Jennifer Rexford, Scott Shenker, Jonathan Turner, CCR 2008.

[5] DevoFlow: Scaling Flow Management for High-performance Networks, Andrew R. Curtis, Jeffrey C. Mogul, Jean Tourrilhes, Praveen Yalagandula, Puneet Sharma, Sujata Banerjee, Sigcomm 2011.

[6] Hedera: Dynamic Flow Scheduling for Data Center Networks, Mohammad Al-Fares, Sivasankar Radhakrishnan, Barath Raghavan, Nelson Huang, Amin Vahdat, NSDI 2010.

[7] ElasticTree: Saving Energy in Data Center Networks, Brandon Heller, Srini Seetharaman, Priya Mahadevan, Yannis Yakoumis, Puneet Sharma, Sujata Banerjee, Nick McKeown, NSDI 2010.

[8] Introduction to Data Mining, Pang-Ning Tan, Michael Stinbach, Vipin Kumar, Pearson, 2008.

[9] A network in a laptop: rapid prototyping for software-defined networks, B Lantz, B Heller, N McKeown – Proceedings of the 9th ACM SIGCOMM Workshop on Hot Topics in Network, Article No. 19, 2010.

[10] A neural network method for minimum delay routing in packet-switched networks, G Feng, C Douligers - Computer Communications, 2001 - Elsevier

[11] A genetic algorithm for shortest path routing problem and the sizing of populations, CW Ahn, RS Ramakrishna - Evolutionary Computation, IEEE …, 2002 - ieeexplore.ieee.org