martes, 15 de octubre de 2013

Review of "Forwarding Metamorphosis: Fast Programmable Match-Action Processing in Hardware for SDN"

  • Did you like this paper? Why? 1-2 sentences.
Its ok for me, because they are promising 2x improvements using a Risc cpu combined with a TCAM changing the architecture. But I'm not pretty sure how it could behave in a real implementation.

Also for me, some parts I could understand and follow the high level idea but some others parts where too much specific.
  • What problem is this paper solving? 2-3 sentences.
They are solving the problem of matching and action, in a switching device using openflow and maintain a high throughput. Specially now that the latest openflow specifications includes reconfigurable match tables. That is very important because will allow a better control of the data plane.  
  • What are the strengths of this paper? 3-4 sentences.
The strengths are the initial improvements and design to building switches allowing match more fields and protocols. 

Also their design if its correct its promising a lot comparing to cost.
  • What are the main weaknesses in the paper? 3-4 sentences.
The main weakness is that is a theorical paper even if its a system paper. But I undersand the cost and amount of work related to this project should be huge and as a first step is convincing to getting funding.
  • What would you do differently? Are there assumptions you disagree with? Do you see ideas for future work or improving the solution proposed? 5-6 sentences
I would ask my self if there is another combination rather than TCAM to improve the design, or for make more convincing the paper a small implementation to demonstrate the concept of the idea works.

jueves, 10 de octubre de 2013

Review of "B4: Experience with a Globally-Deployed Software Defined Wan"

Did you like this paper?
Yes, I liked because they implement a very interesting example of what we could achieve using SDN and the benefits we could get if we keep pushing for this new technology. Also because they show how they solve some problems like the max-min fair allocation. 

What problem is this paper solving?

They are solving some problems that someone would like to implement a SDN network will face, and they give some good details of how they solve them . But their solutions are not the only one and some of them could be optimized.  

What are the strengths of this paper? 3-4 sentences.
The strengths of the paper is learn based on their experience, they did not create a new awesome algorithm instead they implement something very straight forward given the openflow technology available and the constrains related to their applications. Also show how complicate its to combine the components and its an example how could they build in top of others research works like Onix and use the NIB.

What are the main weaknesses in the paper?
They claim that hardware is tend to fail less than software, and they did not compare the performance of their TE Optimization algorithm to the LP solution or any other. 

What would you do differently? Are there assumptions you disagree with? Do you see ideas for future work or improving the solution proposed?

Maybe trying to mix the out-of-band control network with the controlled network, also I may consider the type of application in the traffic split not only considering network throughput also considering network delay. 

jueves, 4 de abril de 2013

Review of "The Akamai Network: A Platform for High-Performance Internet Applications"

Did you like this paper? Why?

Yes, because it briefly explain the problems that the internet is facing and how it affects to organizations and business, for later show the benefits of akamai.

They also explain as a big picture how akamai is designed, and the components related to it, for allowing content delivery.

What problem is this paper solving?
I think they are not solving any problem, they are showing how all the multiples solutions and decisions made by the Akamai designers allow them to get a real business, and having the performance they needed for multiple kind of applications.

What are the strengths of this paper?
I guess that they explain that the collocation of the Edge servers are essential for allowing them to build in top of that all their optimizations (Origin server location, Cache control, Cache indexing) and control, to provide the customer the reliability they want.

What are the main weaknesses in the paper?
They could have explain why is better to use Akamai instead of try to build some special system, to having similar performance. Could be related to price of any other reason, but we don't know.

What would you do differently? Are there assumptions you disagree with? Do you see ideas for future work or improving the solution proposed?

I like the paper, The paper is large enough to understand how is designed Akamai and the problems and benefits it provide, I guess they could explain a little bit about the main problems they are facing on that architecture they build based on their design and how they are planning to improve it.

jueves, 21 de febrero de 2013

Review of "The Click Modular Router"

Did you like this paper? Why?
Yes, I like the paper because click it's a very useful tool for develop experiments. It allows to analyze the incoming network traffic and by our algorithm dispatch the messages as we want to. Also the concept of modules, that have incoming parameters and outputs, allow us to build using someone else components.

What problem is this paper solving?
It is providing a tool "click" for allow us to route packages based on our own algorithms and filters.
Its allow a rapid deployment of small experiments.
Click allow to simulate middleboxes.
Click also have the advantage to allow us to keep a state of the package embedded in the algorithm that we design for routing.

What are the strengths of this paper?
The definition and explanation of the building blocks of click is the mayor contribution of this paper. How it operates and how much performance we could expect.


What are the main weaknesses in the paper?
I guess some work of click is overlap with openflow, both could do some similar things, and allow us to use SDN using a different approach. Openflow is based in rules, Click is based in modules.


What would you do differently? Are there assumptions you disagree with? Do you see ideas for future work or improving the solution proposed?
I would try to simplify the creation of a module, maybe allowing to use another programming language rather than C for having the flexibility of fast development.




martes, 19 de febrero de 2013

Review of "A 50-Gb/s IP Router"

Did you like this paper? Why?
No, because a lot of their design it is based on the CPU, that allow them to achieve that performance, also at the end of their paper they present an incomplete implementation of their idea.
It is true that based on the experiments results they have we can assume they will achieve the desirable performance when they finish the router.

What problem is this paper solving?
They are demonstrating how could be implemented the forwarding engine mainly based on the Alpha 21164 Processor that they mention it is a good solution for their needs. Also they are using a good idea for the allocator algorithm that looks more fare than previous approaches and also it is possible to implement.


What are the strengths of this paper?
I like how they simplify the design of the router backplane compared to McKeown, I guess the most important strengths of the paper its that could be feasible to use an specialized Processor for solve a lot  of the problems related to the backplane.


What are the main weaknesses in the paper?
The allocator algorithm it is not convincing me that will improve a lot the fairness of the system. Could be interesting getting some performance results.


What would you do differently? Are there assumptions you disagree with? Do you see ideas for future work or improving the solution proposed?
I would complete the implementation for having results mention that they are working with the complete implementation of their prototype, and also compare their approach to others routes performance.