Re: [hybi] frame length encoding

[John Tamplin <[e-mail]>]

On Sat, Aug 21, 2010 at 4:57 PM, Shelby Moore <[e-mail]> wrote:

> My understanding is there would be copying, if you read my link again
> carefully:
>
> http://www.ietf.org/mail-archive/web/hybi/current/msg03402.html
>
> The reason is because I understand from what I read at John's links that
> the entire file is contiguous in memory (because it is a memory mapped or
> memory cached copy of the file on the disk), and thus to insert the frame
> headers at the 32bit chunks, you would have to move the entire file
> through the CPU.  Is my understanding incorrect?  Can you simply rewrite
> one sector on disk and the memory manager remaps?


I haven't tried it personally, but I would suspect you could use TCP_CORK,
write a header, use sendfile to send a chunk of the file to finish the
frame, uncork, repeat until done.  I don't think there would be any extra
copying involved here, but you can't just issue the sendfile and go on to
the next message, you need some thread that will queue up the next frame
when it is ready.

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