- From: Amos Jeffries <squid3@treenet.co.nz>
- Date: Sat, 12 Jul 2014 15:42:37 +1200
- To: ietf-http-wg@w3.org
On 12/07/2014 10:37 a.m., Martin Thomson wrote: > On 11 July 2014 14:53, Poul-Henning Kamp <phk@phk.freebsd.dk> wrote: >> If 256 bytes is enough for basic interop, we shouldn't require more. > > I find this reasonably convincing. Interoperability is what we are > looking for here. > > Though I'll note that the arguments regarding cost of implementation > are very much less convincing. Those devices, when presented with a > peer that doesn't know their capabilities, will have to deal with 16k > frames until their setting is seen. And I'm not sure that 16k is so > hard for a constrained device to handle anyway. I keep hearing about > what amazing things that even the most constrained hardware can handle > (TLS stacks, node.js, etc...). So I'm disinclined to lend too much > weight to arguments about limited capabilities. > > That said, if 256 is enough to get something working, then that's a > good argument. > > An alternative approach would be to set the default and minimum to the > same value. That has other benefits, like being able to completely > ignore a setting from the other side. That seems pretty attractive. > When I sueggested that same point some time back you convinced me that is was perfectly reasonable for a constrained device to reject a number too-big frames and streams on connection initiation. Nothing has changed since then. So why is it now such a burden to reject some frames up front on a constrained processor? If the constrained processor is going to pick 256 in the first place its something like an 8-bit processor counting memory by the byte. Even 16-bit ATOM these days have a few hundreds of KB of RAM attached. Amos
Received on Saturday, 12 July 2014 03:43:08 UTC