Re: A rough analysis of the impact of headers on DoS

On 12/07/2014 2:59 p.m., Martin Thomson wrote:
> On 11 July 2014 19:49, Amos Jeffries wrote:
>>
>> HOL blocking a connection while one client incrementally HPACKs streamed
>> headers is orthoginal to HOL blocking how?
> 
> Don't do that.  I think that's been said plenty of times.  If you
> intermediate and multiplex messages from multiple peers onto the same
> connection, you can't safely stream headers.


We must be reading different WG lists. Because the rationale I've been
seeing most strongly behind removing reference set was to increase
streaming of HEADERS+CONTINUATION from A to B without having to worry
about size or buffering. Mostly server peoples arguments, with a case
for not wanting to buffer child application output.
 If one elimitates buffering of headers before send then the HOL is
directly incurred. Nothing orthogonal about that.

>  You need to have an
> entire block before you commit, otherwise you have this problem.  But
> that's *your* fault as the intermediary, not the fault of your clients
> or the protocol.
> 

I do agree completely. Which is why I am neither supporting nor
objecting to the HPACK proposals at this point.

Amos

Received on Saturday, 12 July 2014 05:04:46 UTC