Good question. Yes, for the most part. Specifics vary a little by OS, but
generally it requires a little introspection and to some extent OS-provided
"best guess" projection for the next request/response (this is how to some
extent app developers set constraints on using only wifi, or at least
preventing cellular). We'd need to consciously avoid reaching out when a
new MCC-MNC-to-IP record is clearly out of bounds IPs (e.g., IPs not even
remotely adjacent to known good ones in geo).

-Adam


On Thu, Apr 3, 2014 at 1:28 AM, Yuri Astrakhan <[email protected]>wrote:

> Does the app know if it is connecting over wifi or carrier's network?
> Also, in case of multiple SIM cards, does it know which data connection is
> actually being used?
>
> Thanks!
>
>
> On Wed, Apr 2, 2014 at 9:08 PM, Adam Baso <[email protected]> wrote:
>
>> Hello -
>>
>> I work on the mobile partner engineering team, probably known better as
>> the Wikipedia Zero people.
>>
>> If our servers see that an IP address matches with a given partner mobile
>> network operator (MNO), the pertinent Wikipedia Zero text banner is shown
>> to the mobile web user; additionally, offsite links are rewritten to warn a
>> user when s/he may be entering into a zone that requires data access
>> charges.
>>
>> This is all fine and well, but partner operators sometimes change IP
>> addresses and our systems get out of sync. Our partners are busy and our
>> partner management team does of course strive to work with partners to
>> proactively manage IP and other technical updates, yet inevitably
>> information can fall through the cracks. Consequently, when IPs drift, the
>> system doesn't show banners and do URL rewriting as well as it could.
>>
>> In order to in part more proactively remediate the drift of operator exit
>> IP addresses, we're interested in logging two pieces of information
>> server-side via the forthcoming rewrites of the Wikipedia for Android and
>> Wikipedia for iOS apps, and in some future state a rewritten Firefox OS app:
>>
>> (1) MCC-MNC identification code of operator if present (e.g., 123-45 if
>> the connection is on a particular operator - MCC-MNC is in the format
>> ###-##)
>> (2) Exit IP address (typically, gateway/proxy in MNO infrastructure)
>>
>> The MCC-MNC identification code is embedded on SIM cards and accessible
>> by routine app APIs.
>>
>> We would not want to log this information alongside the other Apache
>> webserver-style elements, but instead have just these two columns in a
>> separate nonpublic file location, purging records older than 90 days.
>>
>> The thought is to just have the app code add the MCC-MNC value to an HTTP
>> header once in a given app session using an MCC-MNC bearing connection
>> (cellular data), and let the server detect the IP as per normal server
>> operation.
>>
>> In a nutshell, after normalizing the MCC/MNC codes (some are likely to be
>> malformed) and cross-checking against our own MCC/MNC database, we'd be
>> able to see if the IPs are askew, and then reach out to operators to ask if
>> they have any updated IPs since the last time we received an official
>> update.
>>
>> We think this is a simple and fairly easy way to observe updated IP
>> addresses for operator partners, and prompt the partner management team to
>> reach out to operators for updated official source IP addresses.
>>
>> The information could incidentally be useful in gauging rough demand for
>> Wikipedia in markets germane to Wikipedia Zero (e.g., higher data costs,
>> lower disposable income), although that is secondary to keeping the IPs
>> accurate.
>>
>> Internal review suggests this is in alignment with privacy policy, and we
>> wanted to see if there were other thoughts on this approach. We plan to
>> move the discussion over to wikitech-l and then later a broader list, but
>> to avoid cross-posting problems with people having one membership in one
>> list but not another, wanted to start on mobile-l first.
>>
>> One last thing - the set of IPs for a given MNO are relatively small. For
>> example, an MNO may have 100 IP addresses representing 1 million cellular
>> subscribers. This has two practical consequences: (1) troublingly, even
>> just a handful of missing IPs has outsize impact, and (2) highly targeted
>> geography and behavioral inferences are unlikely for a data set composed of
>> just two types of data elements submitted.
>>
>> Thanks.
>> -Adam
>>
>> _______________________________________________
>> Mobile-l mailing list
>> [email protected]
>> https://lists.wikimedia.org/mailman/listinfo/mobile-l
>>
>>
>
_______________________________________________
Mobile-l mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/mobile-l

Reply via email to