[Nfd-dev] ndnping timeout at high frequency

[Davide Pesavento]

Yes, you should always use some form of congestion control for a "data
transfer" type of application. The ndncatchunks tool [1] has an
implementation of congestion control (in the aimd and cubic pipeline
types).

[1] https://github.com/named-data/ndn-tools/tree/master/tools/chunks

Davide

On Mon, Jun 10, 2019 at 4:15 PM Ishita Dasgupta
<ishita.dasgupta at gmail.com> wrote:
>
> Hi,
>
> Yes. Apologies, I didn't mention that the connection is over ethernet.
> Let's say this problem transcends to an application (using NDNperf to download data) where the connection simply breaks in certain cases when contention for bandwidth is too high. I'm assuming this could be the same reason causing it? Is the only solution to this problem implementing some sort of congestion control over NDN?
>
> Regards,
> Ishita Dasgupta
>
> On Mon, Jun 10, 2019 at 4:08 PM Davide Pesavento <davidepesa at gmail.com> wrote:
>>
>> Assuming you're using a UDP or Ethernet face between the two nodes, the reason is packet loss due to congestion... NFD can't keep up, socket buffers become full, and the operating system starts dropping packets.
>>
>> Davide
>>
>>
>> On Mon, Jun 10, 2019, 3:42 PM Ishita Dasgupta via Nfd-dev <nfd-dev at lists.cs.ucla.edu> wrote:
>>>
>>> Hi,
>>>
>>> I have two nodes linked with 6 Mbps physical interface. On the producer, I have ndnpingserver running and on the consumer, I run ndnping tests with frequency of 1 ms, 2ms and so on upto 1000ms. I see no issues when I run an ndnping of frequency 3ms or more, however in cases of 2ms frequency or mostly in the case of 1ms, I see upto 25% packet loss due to timeout occuring after a certain number of packet pings. I have monitored that the ndn interests reaches the consumer but the producer doesn't hear back the data packet.
>>> Initially, bandwidth was a concern, but I see that that ndnping data and interest packets are cumulatively 1386 bits. Thus, with 1ms frequency, 1.5 Mbps should be enough (?)
>>> Has anybody else come across this? And knows why the timeout occurs at high frequencies?
>>>
>>> Regards,
>>> Ishita Dasgupta
>>>
>>> _______________________________________________
>>> Nfd-dev mailing list
>>> Nfd-dev at lists.cs.ucla.edu
>>> http://www.lists.cs.ucla.edu/mailman/listinfo/nfd-dev