[Ndn-interest] [Nfd-dev] ndnping

[Davide Pesavento]

Giuseppe,

If you can use NFD (and ndn-cxx) from git master, I suggest to enable
congestion marking in nfd.conf (set "enable_congestion_marking yes")
and use the chunks tool. ndnping doesn't have any congestion control.

Best,
Davide

On Thu, Feb 1, 2018 at 3:15 AM, Giuseppe Carella <gcarella228 at gmail.com> wrote:
> Hello Junxiao,
>
>
> Producer and consumer are not directly connected, because I am using two
> NFDs for the communication.
> They're running on two different hosts.
> I show you in more details the configuration:
>
>
> Producer + NFD <----------------------------------------------------------->
> NFD + Consumer
>
> I set the UDP face and the number of packets lost is huge.
> I tried to set the ether face and it seems that the number of packets lost
> is much higher.
> In TCP, except for the beginning when I get lost some packets, the situation
> is always stable and I never loose packets.
>
>
> 2018-01-31 10:13 GMT+01:00 Giuseppe Carella <gcarella228 at gmail.com>:
>>
>> Hello Junxiao,
>>
>>
>> Producer and consumer are not directly connected, because I am using two
>> NFDs for the communication.
>> They're running on two different hosts.
>> I show you in more details the configuration:
>>
>>
>> Producer + NFD
>> <-----------------------------------------------------------> NFD + Consumer
>>
>> I set the UDP face and the number of packets lost is huge.
>> I tried to set the ether face and it seems that the number of packets lost
>> is much higher.
>> In TCP, except for the beginning when I get lost some packets, the
>> situation is always stable and I never loose packets.
>>
>> Giuseppe.
>>
>>
>>
>>
>> 2018-01-30 15:37 GMT+01:00 Junxiao Shi <shijunxiao at email.arizona.edu>:
>>>
>>> Hi Guiseppe
>>>
>>> What is the Interest interval from each ndnping? `ndnping /app1` without
>>> any other argument gives an Interest interval of 1000ms, i.e. the producer
>>> answers nine Interests per second. At such a low rate, your problem is not
>>> congestion or processing overhead, but somewhere else.
>>> I saw you are using UDP. 6KB is above the default Ethernet MTU, so the IP
>>> stack is going to fragment the IP packet.
>>>
>>> Are your two nodes directly connected? If yes, use Ethernet instead of
>>> UDP.
>>> If there are multiple hops, do you control every link? If yes, increase
>>> MTU so that the Data packets fit in MTU and thus do not require IP
>>> fragmentation.
>>> If you cannot control every link, ask your network administrator to fix
>>> problems in their middleboxes (especially firewalls and NAT routers). An
>>> IP-complaint node is required to forward fragmented IP packets correctly,
>>> and not drop or alter them.
>>>
>>> Yours, Junxiao
>>>
>>> On Tue, Jan 30, 2018 at 8:36 AM, Giuseppe Carella <gcarella228 at gmail.com>
>>> wrote:
>>>>
>>>> Good morning community,
>>>>
>>>> I have modified ndn tools in order to send a 6KB answer for each
>>>> ndnping.
>>>> I have done an experiment, using udp4 face and opening 9 command lines
>>>> from which I
>>>> launched the command "ndnping /app1".
>>>> This is the configuration:
>>>>
>>>> producer(ndnpingserver /app1) <-----------------------------> consumer(9
>>>> ndnping /app1).
>>>>
>>>> What I notice is that a lot of packets are lost. It means that there is
>>>> a problem of congestion in the case NFD receive too many interest packets.
>>>> Do you know if there is a way to configure NFD in order to reduce the
>>>> congestion and loss packets?
>>>> What I have done at the applicative layer to limit this problem is not
>>>> enough unfortunately.
>>>>
>>>>
>>>> Thank you.
>>>> Giuseppe.
>>>
>>>
>>
>
>
> _______________________________________________
> Nfd-dev mailing list
> Nfd-dev at lists.cs.ucla.edu
> http://www.lists.cs.ucla.edu/mailman/listinfo/nfd-dev
>