[DNSOP] Do53 vs DoT vs DoH Page Load Performance Study at ANRW

[Kevin Borgolte <[e-mail]>]

Hi all,

we recently did a study on Do53/DoT/DoH performance on webpage load times in Firefox, which we wanted to share with you. Austin Hounsel will give a talk about our work at ANRW on Monday. Paul Schmitt and I will also be around (myself only Monday afternoon and Tuesday) if you’d like to chat about it in person. The extended pre-print of our paper is available here: https://arxiv.org/abs/1907.08089

We looked at the effect of Do53/DoT/DoH on page load times because we figured they will be a more interesting measure than simple DNS resolution times. We also looked at simulated mobile connections (4G, lossy 4G, and 3G). We found that DoH is on par with Do53/DoT if your Internet connection is good, and worse if it is not. The potential for DoH seems to be largely push. ADD can also make sense, or at least an extended/updated OS API would.

We think there are two opportunities to improve Do53 and DoT: better wire format caching and partial responses combined with multiple questions. Part of the improvements and issues we discovered we later found out (by talking to some of you) were mentioned in early working group drafts, but they never made it into published RFCs or mailing list posts (which is where we looked initially). We hope you can help and explain the thoughts that went into these decisions 15 to 20 years ago. Maybe it even makes sense to revisit the ideas from back then combined with our ideas, considering how the Internet has evolved?

We‘d appreciate any feedback on our work. Please also feel free to reach out to us directly (in person or by email) if you have any insight or feedback you’d prefer not to post to the list.

Best,
Kevin

P.S. Please excuse the posting to multiple lists, but all seem relevant.