Re: [hybi] Flow control quota

[Takeshi Yoshino <[e-mail]>]

Hi Arman,

Thanks for point out these risks.

OK, we should note that the quota need to be replenished to some quantity
without waiting for full consumption, maybe with the example of compression
extension you mentioned.

On Thu, May 31, 2012 at 8:01 PM, Arman Djusupov <[e-mail]> wrote:

> When the per-frame compression extension is being used on a logical
> channel,
> it is impossible to create a frame of exactly the desired size. A few bytes
> of quota will always be unutilized unless some padding is used. The mux
> specification prohibits sending frames bigger than the available quota.
> When
> per-frame compression is applied on the payload before mux, this payload
> cannot be fragmented any more. Even if we make compression extension mux
> aware, it cannot produce a 1 byte compressed frame, it can only flag a
> frame
> as uncompressed and simply send 1 byte. In general, sending 1 byte in a
> separate frame is not desirable.
>
> Extensions that set RSV bits prohibit frame fragmentation. This causes a
> general incompatibility issue with mux FC when an extension is applied
> before the mux. We need to consider whether there is a solution to this.
>

Yes, the multiplexing code need to feed back to extensions behind it to
control fragment size.

Limiting quota to very small value such as 1 byte persistently is not
practical. The bandwidth will be quota / RTT. It's really slow. I think
that warning the risk of dead lock is enough.