[Nfd-dev] Maximum FIB size

[Bruno Ricci]

Hi guys,

first of all, thank you for your answers.

All we did was run a bash script that calls nfdc register. All routes 
are random string of 16 chars, and are set towards a random IP address 
using UDP. As simple as (pseudocode)

for i in 1..100000
do
   ip=random(3.3.3.3)
   name=random(16)
   nfdc register ndn:/$name  udp://$ip
   sleep 1
done

This runs fine without errors (at least with the default debug level), 
until NFD silently crashes. As Junxiao pointed out, maybe the problem 
relies on the number of faces, and not on the FIB size (I honestly 
haven't thought about that, thanks for pointing that out). I will try 
tomorrow morning at work to re run the script reducing the faces number, 
and to run NFD inside gdb.

I will keep you posted, thanks again.

Bruno

On 13/01/2016 19:58, Haowei Yuan wrote:
> Just to add some information about the FIB. We don't have a limit on
> the FIB size. The underlying hash table data structure will double its
> size if the load factor is higher than 50%.
>
> As Susmit and Junxiao mentioned, some more information to reproduce
> the problem would be helpful.
>
> Haowei
>
>
> On Wed, Jan 13, 2016 at 10:39 AM, Junxiao Shi
> <shijunxiao at email.arizona.edu> wrote:
>> Hi Bruno
>>
>> Are you using the same app to register all the routes, or are you using 1000
>> distinct apps?
>> If the latter, this crash could be caused by large number of faces, not FIB.
>>
>> Can you provide a GDB trace for the crash?
>> You'll need a debug build of ndn-cxx and NFD, run NFD within GDB, and type
>> `bt full` (not just `bt`) when crash happens.
>>
>> Yours, Junxiao
>>
>> On Jan 13, 2016 03:30, "Bruno Ricci" <bruno.ricci at uniroma2.it> wrote:
>>>
>>> Hi,
>>>
>>> we're trying to do some experiments about FIB scaling, using NFD out of
>>> git (commit da3ba964301a43f15e6b87c3d585713068252ae6).
>>>
>>> We ran some tests adding random FIB entries, but after adding about 1000
>>> entries (not a constant threshold, the number varies but is always above
>>> 1000) NFD crashes.
>>>
>>> The question then is: is there a maximum number of entries in the FIB?
>>>
>>> Best regards
>>>
>>> Bruno
>>> --
>>> ----------------------------------------------------------------------
>>> Bruno Ricci, Ph.D.
>>> Post-doc Researcher
>>> CNIT - Consorzio Nazionale Interuniversitario per le Telecomunicazioni
>>> Department of Electronic Engineering, University of Rome "Tor Vergata"
>>> Website: http://netgroup.uniroma2.it/bruno-ricci/
>>> Tel.: +39 06 7259 7445
>>> _______________________________________________
>>> Nfd-dev mailing list
>>> Nfd-dev at lists.cs.ucla.edu
>>> http://www.lists.cs.ucla.edu/mailman/listinfo/nfd-dev

-- 
----------------------------------------------------------------------
Bruno Ricci, Ph.D.
Post-doc Researcher
CNIT - Consorzio Nazionale Interuniversitario per le Telecomunicazioni
Department of Electronic Engineering, University of Rome "Tor Vergata"
Website: http://netgroup.uniroma2.it/bruno-ricci/
Tel.: +39 06 7259 7445