c4992735   
   From: none@none.invalid   
      
   On Thu, 3 May 2012 02:21:16 -0700 (PDT), dpcobrien@gmail.com wrote:   
      
   >On Thursday, May 3, 2012 2:15:13 AM UTC+1, Rick Jones wrote:   
   >> Dara   
   >>  wrote:   
   >> > As an exercise I am trying to implement a very simple packet-loss   
   >> > measurement tool (copying the approach in Sting: a TCP-based Network   
   >> > Measurement Tool by Stefan Savage). I am sending a padded HTTP GET   
   >> > request as a sequence of TCP packets with 1 byte of payload to   
   >> > various web servers and watching the acks coming back. I pause for 1   
   >> > second between sending each TCP packet. I have two questions:   
   >>   
   >> I'm not familiar with Sting, but why single-byte segments one second   
   >> apart?   
   >>   
   >> > 1. I am not seeing any packet loss. Am I getting lucky or is the   
   >> > size of the packet (1 byte payload) somehow not big enough to   
   >> > encourage packet loss?   
   >>   
   >> It depends.  Certainly with a single byte segment once a second, you   
   >> are very unlikely to trigger any congestion yourself.  While bit   
   >> errors can lead to dropped segments, 99 times out of 10 (very   
   >> scientific phrase...) a packet is dropped in response to congestion.   
   >>   
   >> I seem to recall some assertions that the "last mile" (as in, to you)   
   >> is most frequently the point of congestion.  So, if you aren't going   
   >> to trigger any congestion in the last mile, the chances of seeing any   
   >> become that much more remote.   
   >   
   >I am trying to measure packet loss on the outbound (rather than inbound)   
   >route. To that end I have been firing packets at web servers around the   
   >globe and hoping to see some packets not making it to their destination   
   >(as evidenced by missing acks). I will try again with bigger packets   
   >and maybe at busier times of the day.   
      
   Missing acks don't guarantee that you had outbound packet loss. Maybe   
   the ack itself got lost, or was just never generated. It seems like   
   your test would be more controlled if you had authority over both end   
   points rather than just one.   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)