Google DNS

There are many very good reasons to be alarmed by the notion that Google would like the world to migrate to their servers for DNS resolution.  A good discussion on the Nanog mailing list:

http://mailman.nanog.org/pipermail/nanog/2009-December/015787.html

highlighted several of these, mainly from the perspective of a network operator or ISP admin, but the usual questions of privacy/tracking/anonymity were raised.

Certainly a monoculture is a dangerous thing, and this makes the sorry state of the global Internet worse in a fairly profound way.

As you may imagine, I have things to say about this. But today, I'd like to submit what I believe to be the impetus behind this new offering:

I believe that Google will begin delivering search results based not only on PageRank but an amalgam of PageRank and other, increasingly "out of band" information.

While a significant portion of my Internet behavior is funneled, in one way or another, through Google search (I don't use mail/voice/groups/etc.) there is a lot of behavior that isn't. This "out of band" behavior is every web site, and especially every chain of web sites that I visit without searching, or after searching for a single site, but then following subsequent links forward.

If search results, and especially ad placement, is generated not just with PageRank, but with a union of PageRank and other information like "site you are most likely to visit within X minutes of these sites" there is the potential of providing even more "relevant" search results, which lead to more "relevant" ad placements, leading to more ad clicks.

This isn't exactly what I would characterize as a "good" in the world, and all of the fears I have heard voiced are well founded. So even though my proposed motivation for this new service is rather benign, the technical underpinning is not any less foreboding.

If this is indeed the motivation for "Google Public DNS", what would this imply ?

First, Google would have an incentive to reduce DNS caching. If you receive name lookups from inside of your own network, or inside of your own system, this decreases the intelligence that is flowing back to Google. It will be very interesting to see what the default cache settings for name resolution are in Google Chrome OS and Android.

Second, the shady world of "search engine optimization" and Adwords fraud is going to get more interesting. These activities could conceivably shift from a passive collection of fake websites and cross-connected "blog entries" to the active generation of network traffic to influence search results.

Which sounds like a real party.

 

Comments

[John Kozubik]

Review of the performance documentation, located here:

http://code.google.com/speed/public-dns/docs/performance.html

neither bolsters nor negates my contention that Google will use their DNS as an adjunct to PageRank, and thus "discourage" caching.

On the one hand, they (rightly) point to "(t)he trend towards lower DNS TTL values" and "cache isolation", but their solution to this is simply to accept these trends and throw more hardware and geographic redundancy at the problem - which is easy, if you're Google.

On the other hand, Google is going to be prefetching name resolution. Their explanation is a bit hazy, but it does not stand to reason that search results will be more relevant based on prefetched name lookups...

So, in addition to examining the default cache settings for name resolution in things like Chrome and Android, I'd also like to see how a "normal" name lookup is formed vs. a prefetched lookup. My contention would imply the ability for Google to distinguish between the two...

[Chris Carlin]

As if the ISPs that are serving the DNS queries are any better?

I'd say at least Google is in a position to do something positive with the data. And in the end, using Google for DNS is a choice that a user could unmake pretty easily if he's not being served well... the default ISPs have far more propensity toward evil with their semi-captive audiences.

I'd say Google's plan is the best of evils in this case.

[John Kozubik]

Well, the point of the above post was not to focus on whether Google Public DNS is a good or a bad thing ... I think it's bad, from the point of view of Internet monoculture, but that's really just an aside ... or a personal comment.

The point here is that I have yet to hear a plausible rationale for the existence of this service, other than a very hazy "google aggregates a lot of data and somehow makes money off of that".

So, the point of the post was what I believe to be a plausible rationale. I think they are going to start to adjunct PageRank with "out of band" data to improve search relevance.

If you want to delve into the infrastructure ramifications of the service, the NANOG link I posted contains a thoughtful discussion.

[Account Deleted]

Your rationale is bogus. Try reading the Privacy page for Google Public DNS. It clearly states:

==
We don't correlate or combine your information from the temporary or permanent logs with any other data that Google might have about your use of other services, such as data from Web Search and data from advertising on the Google content network.
==

[Account Deleted]

Also, I tried reading that NANOG thread, but gave up after 10 or so messages about silly 6.6.6.6/evil not-very-funny jokes :-(.

[John Kozubik]

I don't think the privacy policy in any way precludes Google from using DNS trails to provide "better" search results to the public at large.

If some random web browser shows up and searches for "widgets" and Google has noticed that people who nslookup this widget site tend to also nslookup sites X, Y and Z in some connected fashion afterwords, then Google can adjunct PageRank with that information.

That's not disallowed by the privacy policy. They wouldn't be combining the data with "other data that Google might have about your use of other services" - they would be combining it with PageRank to generate new search results for the public at large.

(never mind how much trust we should place in their privacy policy - let's assume for the sake of discussion that it is never diluted and always enforced)

[Scot]

I think the actual reason behind Google's move is that it should make Chrome (and by extension, Chrome OS) slightly faster.

http://blog.chromium.org/2009/12/technically-speaking-what-makes-google.html

If Chrome is doing DNS pre-resolution to make things seem faster then they probably want to make sure that DNS requests are handled as quickly as possible. My bet is that Google figured that the best way to speed up DNS was to host it's own DNS service.

I have to admit, when I switched over to Google's DNS Chrome was ever so slightly faster (though Firefox 3.0 remained about the same). While it's only a slight improvement, Google is trying to make the fastest browser and every slight improvement helps.