Re: #40 - HTTP2 default value for client supported max_concurrent_streams

[Roberto Peon <[e-mail]>]

That seems like it might be backwards? The *average* number of requests for
a page is ~70 in 2011 and close to 90 today. This isn't just about a few
instances of demanding pages with many resources. Those pages probably use
in excess of 100 (and Patrick would like to see a larger default than that,
something which the trends support...).

The cheap chip can handle the case where the clients makes too many
requests by rejecting all but one stream, and causing the browser to retry.
In such cases, latency must not matter, else the device would have enough
memory to handle more than one request at once. In any case, given that
mostly we access these minimal devices on the LAN, latency will be minimal
anyway regardless of what happens in the first exchange.

The sweet spot here for defaults is such that it allows for a latency
decrease on high RTT  links as compared to HTTP/1, and such that we we
don't mask malicious behavior.
 If you set the default too low you end up providing a latency bias to
larger sites, which seems unfair to me, and also undermines reasons for
adoption.

-=R