Re: [hybi] Multiplexing extension spec draft 02

[Takeshi Yoshino <[e-mail]>]

Hi,

On Fri, Jan 27, 2012 at 15:57, Martin Sustrik <[e-mail]> wrote:

> On 01/27/2012 01:54 PM, Takeshi Yoshino wrote:
>
>> http://tools.ietf.org/html/**draft-tamplin-hybi-google-mux-**02<http://tools.ietf.org/html/draft-tamplin-hybi-google-mux-02>
>>
>
> I would say it's worth explicitly stating that an endpoint MUST NOT
> announce larger quota than it can immediately read from the network (i.e.
> the amount of empty space in channel's rx buffer).
>

I agree it's a caveat good to note, but we shouldn't limit flexibility
of scheduling granularity by putting such MUST NOT in the protocol spec, I
think. There might be some case where one wants to give more slot to some
particular logical channel greater than size of that channel's buffer (any
demand of large time slot for a particular channel, or any special QoS
algorithm).


> Failure to do so can result in all the channels being stuck because of the
> single offending channel.


I'd add this generalized implementation note if necessary.

Be careful when you assign big quota for some logical channel. There is
risk that the quota exceeds the processing capacity (buffer size,
processing power, bandwidth, etc.) for that channel, and the endpoint can
get stuck by getting excess data for that single offending channel
occupying one shared TCP connection while there's room of processing
capacity for the other channels.