Re: #445: Transfer-codings

I wouldn't mind fixing range-requests, but I don't believe that range
requests make a whole lot of sense for dynamic objects anyway.
I'm more than happy to have my ignorance corrected :)


On Sun, Apr 6, 2014 at 10:29 PM, Matthew Kerwin <matthew@kerwin.net.au>wrote:

> On 7 April 2014 15:04, Roberto Peon <grmocg@gmail.com> wrote:
>
>>
>> I still don't think compression at the protocol/stream layer makes sense.
>> In my experience, it never worked well in our (SPDY) experiments: it added
>> complexity, pened the door for lots of DoS vulnerabilities, increased
>> memory requirements, increased CPU requirements, and rarely helped w.r.t.
>> bandwidth for
>> well-constructed sites which compressed their resources.
>>
>> The cost/benefit here is extremely dubious.
>>
>>
> As I've said, I'm happy to not block HTTP/2 on this, if we can address it
> in HTTP/3, and if H3 isn't going to be another 15 years down the track.
>

I wouldn't want it then, either, honestly. I that that having random
intermediaries attempt to apply a transformation to the content that the
origin wouldn't seems like a mistake.


>
> But I have to respond to one particular type of comment that keeps coming
> up: "*well-constructed sites which compressed their resources.*"
>
> That's a big value judgement on what constitutes good site design. Yes, in
> lots of cases it makes sense to compress your resources and have multiple
> representations, especially for static resources; but what about the sites
> that aren't like that? Why is it bad site design to have a big resource
> that can be accessed with ranges?
>

At best it is wasteful to have to generate a large dynamic resource, and
then serve only a portion of it.
At worst, when the dynamic resource isn't properly versioned, it is just
plain wrong.



> If the answer is because doing so would require TE in order to have
> compression, then it's a tautology (TE is only needed by bad sites; those
> sites are bad because they need TE). If it's because caches don't handle
> that properly, then it's a chicken-and-egg problem. The only reasons I can
> think of for calling them bad are either circular, or "I don't like them."
> Is there a real reason?
>
> Why should I make my web API use "?start=N&end=M" when I could use "Range:
> x-records=N-M" ?
>

It works better for caches, which can (with little logic) cache the chunks
which are requested more often, and/or purge unneeded resources more
easily/naively, which increases cache hit rate for other resources.
In the case of video/audio, it allows for quality/encoding changes while
still allowing for caching.

-=R


>
>
> --
>   Matthew Kerwin
>   http://matthew.kerwin.net.au/
>