===========================================================================
                              CCR Review #43E
                 Updated Tuesday 23 Mar 2010 4:37:12pm EDT
---------------------------------------------------------------------------
   Paper #43: Investigating the Impact of Service Provider NAT on
              Residential Broadband Users
---------------------------------------------------------------------------

                         Timeliness: 2. Thoroughly understood area with
                                        little new work
                            Novelty: 1. Little to add to existing work
              Technical correctness: 5. No errors
                            Clarity: 5. Lucid
                     Recommendation: 3. Reject

                    ===== Summary of contribution =====

The paper presents an analysis of packet traces at a residential ISP that has 
not deployed SPNAT. The traces are used to examine TCP and UDP flow rates to 
understand how SPNATs would have to be provisioned. The paper proposes a UDP 
timeout value that reduces the amount of session state that the SPNAT would 
have to keep track of.

                       ===== Detailed comments =====

I'm concerned that the paper's findings have a very short half-life. As new 
applications become popular, the behavior of number of connections that a PC 
will open will change. BitTorrent was a great example of that. IPTV is a new 
upcoming app that could change this. Gaming consoles also have an impact.

The paper is contributing only a little bit on top of existing residential 
trace analysis such as [12],[15], with [12] being the closest.

I was dissappointed that the paper did not analyze inbound traffic. That is a 
challenging problem for SPNATs and would have been more interesting than the 
analysis of outbound traffic.

The paper needs to clearly identify the trace duration. The dates are given, 
but not the times. So I don't know whether it is 5 full days or really 4 or 3 
full days.

The paper needs to give me a sense for how comprehensive the trace is. I can 
understand that the number of customers cannot be revealed. But can you reveal 
how large of a geographic area the trace covers? Or what the population 
coverage is (which this ISP may not cover completely due to competitors)?

The rather large number of concurrent TCP sessions is troubling. Why does this 
happen? Is there a business customer in there that has a lot of employees 
behind a single IP?

The UDP analysis didn't say anything about DNS. How many were DNS?

The paper is very well written and is easy to understand. I appreciated the 
details in the methodology of how the traces were handled.

Typo in last paragraph of Sec3 -- "this section" --> "next section"