Re: [hybi] frame length encoding

[John Tamplin <[e-mail]>]

On Sat, Aug 21, 2010 at 2:39 PM, Daniel Stenberg <[e-mail]> wrote:

> On Fri, 20 Aug 2010, John Tamplin wrote:
>
>  I thought I would ask for a "show of hands" about three possibilities.  If
>> you feel strongly for or against a particular option, please reply and say
>> so.  If you are fine with any of the three, then there is no need to reply.
>>
>
> I think I would prefer a plain 32 bit size, with no special treatment for
> smaller fragments.
>

See previous discussions -- many people thought it was unacceptable to have
to fragment a message which was already entirely in memory or a disk file
which could be sent with sendfile.  My conclusion from that discussion was
that there was no way we were going to achieve consensus with anything less
than a very large maximum frame size.

Aside from that, your suggestion more than doubles the frame size for small
frames.  As previously indicated, I think that is important because traces
of real-world WebSocket apps show that once compression is added, 98-99.9%
of frames will fit in approximately 126 bytes.  Paying three extra bytes for
the most common frames in exchange for a small gain in simplicity seems a
poor trade-off.

Combining these requirements seems to indicate that some form of encoding
the length in a variable number of bytes is required -- we are debating the
best way to do so.

-- 
John A. Tamplin
Software Engineer (GWT), Google