COM-208: DNS servers inefficiency

Hello,

I am wondering if the DNS system is not a bit inefficient if we want to get a resource on the internet.
If for instance you want to go to the "epfl.ch" website, what happens is that the browser first translate "epfl.ch" to an IP address (using the DNS servers) and then does a request to that webserver for the web page.

Coudn't those requests be merged, so that instead of doing two strips, we would only need to do one ?
So when we type "epfl.ch" in the web browser, it would use the DNS servers to find the authoritative server of epfl, and that authoritative server instead of replying with the IP address, it would directly reply with the web page ?

Re: DNS servers inefficiency

by Katerina Argyraki - Monday, 25 October 2021, 12:36

This is an awesome question, Vladimir! Indeed, one might argue that decoupling (separating) DNS from other services is inefficient. One reason for this decoupling is historical: DNS was not there from the beginning, it showed up at some point because of need (check https://www.cs.cornell.edu/people/egs/615/mockapetris.pdf if interested). Another reason is that this inefficiency was not deemed significant enough for adding more load/complexity to the DNS system: DNS caching works well (most people tend to visit the same servers) and hides much of the inefficiency. That said, there have been research proposals through the years to integrate DNS requests in the Internet architecture in a more seamless way, such that end-systems do not need to make first a DNS request and then a separate service (e.g., web) request. E.g., https://apps.dtic.mil/sti/pdfs/ADA414840.pdf (warning: badly written, hard to read -- but interesting).