=========================
= Peer Review - Round 2 =
=                       =
= "Sorting Networks"    =
= Maria Aguilar R.      =                                          =
= Namratha Sanjay       =
=========================

I won't hesitate to say that this is a significant improvement over the previous version! Well done!
Everything is much clearer, both language and content-wise. 

I consequently have significantly less to comment on :)

In summary, it needs a little more work, but you're pretty much there, I think.
Proofread some of the unclear sections, find sources for some of your statements and perhaps give a little 
more explanation/motivation for some of the concepts that you mention. Since this is a survey, a brief overview 
of why it is relevant would be enough, I think.

Same as last time, I've left comments directly in the pdf. Some are quite long and it's not obvious if the whole 
comment is visible when you hover over it, so make sure to click them :)

-J

=========================
= Comments              =
=========================

INTRO:
======
* The computational requirements are a bit clearer now: Rearrangement of data using comparisons in which hardware and algorithm need to cooperate in an optimal way.

MAIN TEXT:
==========
* You haven't shown the working out for your roofline model. The numbers don't seem to correspond to that of Table 1, or anything else in the survey itself.
  It makes it very hard to critique, in any case.
* The most computationally expensive components are not really identified. Though there is mention of the input size having an effect.
* It's good that you've reduced your scope to CPU & GPU, though it is a shame that FPGAs were dropped...
* Some ideas that were mentioned in the previous version are gone (eg.bandwidth independent sorting), but the remaining ones are not really explained in more detail... It still feels overall as though some concepts are lacking explanation/motivation, especially in the conclusion.. 
* You say you are going to discuss energy efficiency tradeoffs, but then don't.
* The design proposal is short, but sounds plausible in the sense that you've identified that sorting is mem-bound, and are proposing ways to reduce this.
* The programmability of CPUs/GPUs when it comes to sorting algorithms is not really mentioned. It is mentioned that algorithms can be adapted for different platforms, but detail is lacking.
* Your bibliography is a much more reasonable length now. :)

GENERAL:
========
* The presentation is good.
* The language is vastly improved! There are still a few unclear sections though.