Re: [Doh] [dns-privacy] [DNSOP] Do53 vs DoT vs DoH Page Load Performance Study at ANRW

Thanks for sharing the results of your work. It will be great to have
the software available so others can run the experiments from other
locations.

When looking at the page load results the CDF graphs comparing the
various services are very useful to see the relative performance of
different services. However I could not find the range of time values
for the page loads in the experiment. Basically what percentage of the
page load time variation was related to DNS?

Note: For Google DoH, we will be reviewing our implementation for
latency. BTW we have launched our production RFC 8484 DoH service
recently at https://dns.google/dns-query
(https://security.googleblog.com/2019/06/google-public-dns-over-https-doh.html).
It will be great if you can update your software to use this endpoint.

* The experiment was run from Princeton, New Jersey in Northeast US.
The location is in a very well connected part of the world between
network peering points in NYC and Washington DC. You will not see much
difference (due to network latency) between the cloud providers and
the default (local) Do53. Running the experiment from locations which
are further away from cloud providers would provide another
interesting set of data.

* Conclusion on benefit (or lack) of ECS.
Did the page load measurements include content that would benefit from
proximity to the end user, e.g. streaming videos or large downloads?
This kind of content benefits from ECS when the resolver is further
away from the client.

Thanks,
Puneet

On Fri, Jul 19, 2019 at 1:42 AM Kevin Borgolte <kevin@iseclab.org> wrote:
>
>
> > This paper looks interesting. Is the software used in the paper published?
>
> Thanks! The code isn’t open source yet, but we will make it public alongside the Docker setup we used for running it. Not sure when that is going to happen exactly though.
>
> > Or, at least, is the test page set published? I haven't read the whole thing yet, but it seems like the page set would be relevant if the paper tests page load time.
>
> The list of websites is attached. It is extracted from the top 1,000 and 99,000 to 100,000 of a Tranco list.
>
> Best,
> Kevin
>
> _______________________________________________
> dns-privacy mailing list
> dns-privacy@ietf.org
> https://www.ietf.org/mailman/listinfo/dns-privacy