Re: [WARNING: ATTACHMENT(S) MAY CONTAIN MALWARE]Re: HTTP(3) priorities

Hi Oliver,

Thank you for your comments.

Being a server developer, I like the idea of splitting the urgency
groups, but I am not sure if we should bake something that we *might*
want to use into the core design. IIUC, your suggestion is to have a
more fine-grained signaling between the web application and the H2/H3
terminator. Assuming that is the case, these two parties can do an
experiment, by defining a proprietary parameter of the Proxy header
field, do the experiments, then ask for it to become an official
extension.

Now that we are dropping support for priorities in H3, it is
beneficial to have some alternative shipped at an early moment.

Therefore, it is my view that we should first standardize something
minimally viable, based on what the clients and servers already do, at
the same time leaving room for experiments and improvements. While I
agree that having 20*8 urgency levels is possible, it is complex than
just having 8 urgency levels. Broad adoption of the H2 prioritization
scheme happened, even though it was implementable. At least two
large-scale server operators do support it the way spec is written.
But for others, it seemed too complicated. I think we would like to
minimize the chance of repeating that failure.

Regarding your other comments, I am not sure if I am following, but I
do not think that routers or switches would deal with the HTTP-level
priority information. Plaintext H2 is not a thing, and H3 will always
be encrypted. It is also my understanding that sending some packets of
a connection with a different QoS label is a bad idea, because doing
so would have bad impact on loss recovery and congestion control. I
could be wrong (or there might be some solution that I am not aware of
- I'm not a transport persone), though.

2019年7月25日(木) 2:48 Oliver, Wesley, Vodacom South Africa (External)
<Wesley.Oliver@vcontractor.co.za>:
>
> Hi,
>
> The only thing I could suggest is that you make those priority flags increments of 20,
> So that if there are changes or minor priority preference per website, then
> That in between numbers can be used. If there is any new future priority conventions
> Then just add them in the middle of the 20 group splitting the existing group.
> Maybe make the field 2 bytes, which allows for future expansion, were there are many files, streams
> Which require a new priority order.
>
> Otherwise one needs to build a priority map, over the deps, converting a high resolution to low resolution.
>
> Kind Regards,
>
> Wesley Oliver
>
> > On 24 Jul 2019, at 19:14, Kazuho Oku <kazuhooku@gmail.com> wrote:
> >
> > Hi!
> >
> > Attached are the slides that Lucas and I presented during the side meeting.
> >
> > Thank you all for the feedback.
> >
> > 2019年7月9日(火) 21:00 Ian Swett <ianswett=40google.com@dmarc.ietf.org>:
> >>
> >> There has been quite a bit of discussion of HTTP priorities lately on both the QUIC and HTTP mailing lists, with more to follow.  The plan for IETF in Montreal is as follows:
> >>
> >> Monday: Overview of priorities for 15 minutes in HTTP, with 15 minutes discussion.
> >> Wednesday: A side meeting will be held from 8:30-9:45am in Van Horne.
> >> Thursday: I'll summarise the side meeting in Wednesday's HTTP session.
> >>
> >> Thanks, Ian
> >
> >
> >
> > --
> > Kazuho Oku
> > <The Priority Header Field.pdf>
>
> This e-mail is classified C2 - Vodacom Restricted - Information to be used inside Vodacom but it may be shared with authorised partners.
>
> "This e-mail is sent on the Terms and Conditions that can be accessed by Clicking on this linkhttps://www.vodacom.co.za/vodacom/terms/email-acceptable-user-policy"



-- 
Kazuho Oku

Received on Friday, 26 July 2019 13:48:58 UTC