[dns-operations] Writeup of Spring Workshop

[Ralf Weber]

Moin!

On 12 May 2015, at 19:09, Doug Barton wrote:

> On 5/12/15 9:52 AM, Rick Jones wrote:
>> To what extent does the broadening availability of TCP Fast Open 
>> affect
>> the cost of TCP versus UDP calculus?
>
> Yes, that's similar to what I was going to reply to Ralf, more 
> generally if TCP performance were better for DNS (i.e., it's less of a 
> "cost" to the system) does the inevitable fallback to TCP for real DNS 
> traffic become more palatable for you?
Not much. The amount of round trips is only one part of the problem. The 
maybe bigger one is the amount of state that TCP introduces into your 
code, which requires memory and synchronisation which slows the server 
down considerably.

> It's fine to say "EDNS will solve this," but 15 years later it hasn't 
> yet. :)
I haven't done measurements, but I assume there is more EDNS deployed 
for DNS than TCP fast open.

So long
-Ralf
---
Ralf Weber