I remember reading a paper a while ago that showed that if you send two consecutive UDP packets with exact same data over the internet, at least 1 of them will arrive to the destination at pretty high success rate (something like 99.99%)
I wonder if this still works with current internet infrastructure, and if this trick is still used in real-time streaming protocols.
Generally you'd just implement a more generic FEC algorithm though unless you had 2 separate paths you wanted to try (e.g. race a cable modem and 4G with every packet and if one side drops it hope the other side still finishes the race) as there are FEC options that allow non integer redundancy levels and can reduce header overhead compared to sending multiple copies of small packets.
Not per se. The drop chance for consecutive packets is likely correlated, such that if you know the first one was dropped you should increase your prior that the second one will also be dropped.
On the other hand if the root question is if there is an outage style issue then yeah if the path to the destination is having a hard down style issue no number of packets are going to help because they are all going to drop. Likewise if the question is "on a short enough time scale is reliability of delivering a single packet somewhere on the internet ever less than 99%" then yeah somewhere there is a failure scenario and if you look at a short enough time scale any failure scenario can be made to say there is 0% reliability.
> Also keep in mind this note: http://technet.microsoft.com/en-us/library/cc940021.aspx
> Basically, if you send() 2 or more UDP datagrams in quick succession, and the OS has to resolve the destination with ARP, all but the 1 packet is dropped until you get an ARP reply (this behavior isn't entirely unique to windows, btw).
If you are streaming a live sports event you don't want the video frames to fall behind from the real clock too far, but you also don't want the video to stutter when there is packet loss. I believe you can use the UDP trick I mentioned above to amortize loss of high quality frame data, with "double sent" low quality frame data.
Not necessarily. I am not an expert on this topic, but I believe physical device failure is one of the least common reasons a packet is dropped (at least on physical cable transmission). I think link saturation, priority routing (for certain types of packets, or contractual reasons) might be more common. There are thousands of routers out there with a variety of different configurations that decide which packets go through under which conditions.
Can someone summarize the problem in abstract terms? What are the parameters that the algorithm can play with? Packet size and rate of transmission, anything else?
I guess I meant it was an interesting topic to me when I read the paper :) It might not be that exciting overall.