Re: [quicwg/base-drafts] Changing the Default QUIC ACK Policy (#3529)

[Gorry Fairhurst <[e-mail]>]

There are some different points of pain here, the default and the need 
to adapt to the endpoint:

An ACK Ratio (AR 1:1): is the finest granuality of feedback. A QUIC AR 
of 1:1 would consume return path capacity proportional to about 6% of 
the forward traffic (assuming no loss, etc). This is more than TCP 
(about 3%).

An ACK Ratio (AR 1:2): For a return path bottleneck, QUIC with an AR of 
1:2 will still have more impact on sharing flows than TCP would with the 
same AR 1:2 (because QUIC ACKs are larger, and because networks do not 
currently thin QUIC ACKs).

So QUIC needs to choose the default  - is it 1:2, 1:4, 1:8, 1:10?

A QUIC AR 1:10 results in an ACK traffic volume of about 1% of the 
forward traffic. To me, that's a small proportion, which is why I argued 
that 1:10 is a good recommendation. The default is important when: the 
endpoints have no knowledge of the asymetry or the impact they have on a 
link's layer 2 loading, or the endpoints need to reduce the collateral 
damage to other IP flows sharing the return path capacity.

Negotiation: The AR can be driven by the application or endpoint - 
e.g.,  a higher rate connection may well have a benefit from an AR of 
1:10, 1:20, 1:100,  etc - but such ARs are seldom chosen to reduce the 
impact on sharing a return bottleneck, more likely they are chosen to 
benefit the performance of an application or endpoint. That 
consideration is different, and likely depends on the application or 
ednpoint - which to me is where the flexibility offered by a transport 
parameter is so important. draft-iyengar-quic-delayed-ack will still 
need guidance on the "default".

I argued that the default policy impacts the congestion behaviour when 
there is appreciable asymmetry this causes at layer 2 or the collateral 
damage (congestion impact on sharing flows). Sure, the choice of 1% is 
just a number I said was reasonable, but I've seldom encountered L2 
technologies where less than 1% of forward traffic would be 
significantly better.

Gorry


On 27/05/2020 10:31, Bob Briscoe wrote:
>
> To be clear, when I said that having a default policy is not a matter of
> interoperability, that's only strictly true if QUIC supports something
> like draft-iyengar-quic-delayed-ack for the sender to ask the receiver
> to use a certain policy.
>
>
> Bob
>
> -- 
> ________________________________________________________________
> Bob Briscoe http://bobbriscoe.net/
>
> —
> You are receiving this because you were mentioned.
> Reply to this email directly, view it on GitHub 
> <https://github.com/quicwg/base-drafts/issues/3529#issuecomment-634544299>, 
> or unsubscribe 
> <https://github.com/notifications/unsubscribe-auth/ABYLLERFGWEHGO2FSSOAQFDRTTMXTANCNFSM4LOJ4RQA>.
>
-- 
G. Fairhurst, School of Engineering



-- 
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/3529#issuecomment-635379450