Re: [quicwg/base-drafts] DATA frame encoding is inefficient for long dynamically generated bodies (#1885)

[Dmitri Tikhonov <[e-mail]>]

@LPardue writes:

> I'm sympathetic to the issue but not sure it needs fixing in v1. What is worse case overhead?

When one transfers large files, the overhead incurred by the DATA frames will be on the order of 3 bytes for every `[ ~1 MSS, 2^14 - 1 ]` bytes of content, or between about 0.2% and 0.02%.

@ianswett writes:

> The only two cases I can think of where that wouldn't be true are server push, which is rarely used, and trailers, which are rarely supported and even more rarely used.

The server push will continue to be rarely used if the API makes its use inconvenient.  For example, one can imagine that a server may want to push after sending a few KB of data (a finite DATA frame) -- to parse the HTML for CSS links or what-not.  An implementation that only allows pushes before the first (and only) DATA frame will make using server push in such manner more difficult.  Because it is easier to support only a single DATA frame of indeterminate size, it is likely that many implementations will do just that.

The complexity associated with framing data into DATA frames is real (if you want to do it efficiently), but not insurmountable.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/1885#issuecomment-431348155