Re: Proposal - Reduce HTTP2 frame length from 16 to 12 bits

[Leif Hedstrom <[e-mail]>]

On 5/28/13 7:12 PM, Roberto Peon wrote:
> I understand what you're saying. We're all agreeing that the max frame 
> size should be limited. I'm simply suggesting 16k is what I think is the 
> best small frame size. Patrick is suggesting 4k.
> I wouldn't object strenuously to 4k, but my experience and gut both say 
> 16k is the better choice.
>

My gut says that 4k is on the small side too. If we left it at 16-bit, but 
with a SHOULD saying only 14-bits are to be used, are we still cutting too 
much rope for people to hang themselves? I'm not suggesting UA developers 
should have to try to optimize for anything beyond the suggested 16KB frame 
sizes, that seems unreasonable and unnecessary.

On the argument side for smaller frames, CPUs tend to get more powerful 
faster than the NIC capacity increases. In my experience, a modern box can 
saturate a 10GigE at the same (small) object sizes as a machine could 
saturate a 1 GigE NIC 4-5 years ago.

Cheers,

-- Leif