Re: [Idnet] Intelligence-Defined Network Architecture and Call for Interests

[Lori Jakab <[e-mail]>]

On 3/31/17 7:57 AM, Jérôme François wrote:
> Hi,
>
> Data is primordial but sharing them is very difficult (I'm not an
> operator) due to privacy but also legislative issues.
> Here are few comments:
> 1) on anonymization: this far from being simple and I do not believe
> that anybody can guarentee that nothing will be leaked out as
> sensitive data (and at the end this will the responsability of the
> operator who have shared the data).
> -> That is why sharing but not in public manner (with a NDA) is still
> the most viable solution (in my opinion)
>
> 2) charaterizing the data we would like to access is clearly a first
> step (even if many of us like to have raw data)
>
> 3) Another possibility is to run the algorithms you want to test on a
> remote platform without accessing the data and get back the results
> (also here there might be some safguards to avoid that program tries
> to extract sensitive data as results !)

I think 3) is probably the most viable way forward for now. Lack of
public data sets is not a new challenge for networking research, and I
doubt saying that we're going to use traffic traces for ML applications
is going to change that. Creating an open source platform, either pure
software or software + hardware (like RIPE Atlas probes) that runs the
algorithms on the data sending back only the results may have a better
chance of success.

-Lori

>
> jerome
>
> Le 31/03/2017 à 06:15, Laurent Ciavaglia a écrit :
>> Also CAIDA has multiple measurement tools to collect data.
>> Some IETFers are specialists of these tools and report findings on
>> data collected via these tools.
>>
>> Might be something to look at.
>>
>> Also, and with this I'm finished for today, documenting the
>> characteristics of the requested data set(s) might help.
>>
>> BR, Laurent.
>>
>> On 31/03/2017 06:12, Laurent Ciavaglia wrote:
>>> Hello,
>>>
>>> A couple of suggestions:
>>>     -Why not raise this point / request in various WGs (e.g. in the
>>> OPS Area) and RGs (e.g. MAPRG, NMRG).
>>>     -Why not work with the IETF meeting NOC to collect data sets?
>>>
>>> BR, Laurent.
>>>
>>> On 31/03/2017 05:07, Pedro Martinez-Julia wrote:
>>>> Hi all,
>>>>
>>>> That's it and I insist that, if anybody in the mailing list is bound to
>>>> an operator or a similar organization that manages huge amounts of data,
>>>> just the data we need, please try to provide it. We will be able to help
>>>> you to build some anonymization solution, if needed. Thank you.
>>>>
>>>> Regards,
>>>> Pedro
>>>>
>>>
>>> -- 
>>>
>>> Laurent Ciavaglia
>>>
>>> Nokia, Bell Labs
>>>
>>>  
>>>
>>> [phone]
>>>
>>> route de Villejust - Nozay, France
>>>
>>> linkedin.com/in/laurent.ciavaglia
>>>
>>
>> -- 
>>
>> Laurent Ciavaglia
>>
>> Nokia, Bell Labs
>>
>>  
>>
>> [phone]
>>
>> route de Villejust - Nozay, France
>>
>> linkedin.com/in/laurent.ciavaglia
>>
>>
>>
>> _______________________________________________
>> IDNET mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/idnet
>
>
>
> _______________________________________________
> IDNET mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/idnet