Re: #282: Recommend minimum sizes for protocol elements

On 22/06/2011, at 5:03 PM, Willy Tarreau wrote:

> On Wed, Jun 22, 2011 at 06:35:21AM +0000, Poul-Henning Kamp wrote:
>> In message <FEEB46BC-14A2-4131-9309-584EA8813358@mnot.net>, Mark Nottingham wri
>> tes:
>> 
>>> Again -- this is NOT recommending how large people should make cookies, 
>>> but recommending a floor for implementations to support, to improve 
>>> interop.
>> 
>> I agree, but the floor should not be set punishingly high to cater
>> for clueless people.
>> 
>> Standards should promote interoperability, not stupid behaviour.
> 
> Indeed. My observations in field is that clueless people justify their
> stupid designs by "but look, it's permitted". Till now I've only been
> able to show them they were doing stupid things by giving examples of
> various implementations' limits, for instance by reminding them that
> the ubiquitous Apache server had a 8kB limit per header and that that
> should ring a bell in the guy's head.
> 
> Also Mark, I agree the Alteon would be faulty for 1.5kB right now, but
> it was 10 years ago (WebOS 8). With WebOS 10 one year later, they
> increased the limit to 4.5kB. But seeing that people were already able
> to send about 2kB of cookies 10 years ago when DSL was still rare, we
> surely can imagine what they'll do today if the standard suggests that
> everything in the path should be able to support at least 20kB.


Understood. I'd also like to not have to revise HTTP again in another ten years :)

I think 20k made sense to me because of my experiences deploying proxies; however I agree we shouldn't be encouraging large headers. 

Anyone else with opinions?

--
Mark Nottingham   http://www.mnot.net/

Received on Wednesday, 22 June 2011 09:25:00 UTC