Re: [perpass] HTTP user-agent fingerprinting

On Sep 13, 2013, at 3:55 PM, Poul-Henning Kamp <phk@phk.freebsd.dk> wrote:

> In message <A5264783-BDBC-445B-B97D-0764319FC38C@la-grange.net>, Karl Dubost wr
> ites:
> 
>> Note that anything which is removed will break something. 
> 
> Yes, undoubtedly.
> 
> But if HTTP/2.0 limits User-Agent to 32 bytes, that means that
> no HTTP/2.0 browser will send more than 32 bytes, and no website
> will support HTTP/2.0 until they can do their job with just 32 bytes
> of User-Agent.

I imagine that for some time during HTTP/2 early deployments, the primary server technology would include HTTP/2.0 -> HTTP/1.1 proxies. As such, those same HTTP/1.1 servers would presumably still require the same UA fingerprinting as they do today.

From prior experiences, most IPv6 deployments I've done used v6tov4 proxies. We would roll v6 out on a small number of servers, yet still provide the feature for all existing IPv4 services. It seems reasonable to think that HTTP/2 would be deployed (initially) in a similar fashion?

Yes, I know, the HTTP/2->HTTP/1.1 proxies could rewrite the UA. :) I probably would if we changed the UA requirements (e.g. 32-byte size limit).

Cheers,

-- Leif

Received on Monday, 16 September 2013 15:01:46 UTC