[Bloat] [NNagain] The Verge: The quiet plan to make the internet feel faster

Tue Dec 12 05:02:56 EST 2023

Hi Frantisek,

> On Dec 11, 2023, at 20:33, Frantisek Borsik <frantisek.borsik at gmail.com> wrote:
> 
> Hello Sebastian,
> 
> So you have shared that "L$S bible"  :-)

	I even have some sympathy for his stance, by all means, flow rate fairness is rarely optimal and with sufficient information it seems almost always "easy" to do better. It is just that I consider the "with sufficient information" part the tricky bit. A random node in the internet can reliably use 4-tuples to segregate "flow-aggregates"* and arbitrate/equalize capacity between them, which while rarely optimal will almost guarantee that no flow is going to be starved (either none or all ;) ) in a sense is is not the most optimal, but the least pessimal policy. Anything else requires robust and reliable information delivery to the nodes having to make decisions which might work well in controlled environment, but not realistically in the open internet. And just to en-passant kill the "cost fairness" idea pushed in the paper, that requires a "notion" of user over which to aggregate and equalize the cost metric, something an arbitrary node will not have...

TO cite: "	• Allocate congestion among the bits sent by economic entities (cost fairness)" good luck for an intermediate node to aggregate by economic entities in a robust and reliable fashion...

At which point we can safely put that specific idea to rest by virtue of being unrealistic... 

*) This is relatively save to do, as these same tuples are also used in making routing decisions so are not all that easy to game arbitrarily and hence require little to no trust from the intermedate node basing decisions on this information.

> with the bloat list and I do believe that even people at NN list should read it: https://pbg.cs.illinois.edu/courses/cs598fa09/readings/b07.pdf
> 
> Heck, everyone in the field should read it. Food for thought, certainly.

	Respectfully, I think nobody should have to read that (unless they want to of course), it is way too long and misleading given the fact that it is a phantasy, not a realistic solution for the internet. Also I have been using flow queueing schedulers now for roughly a decade and can from that experience state, this solution is "good enough" for decent internet experience and simply works (at least on my link for my traffic mix). Again, I keep my response of NNagain, as I still am not convinces folks over there want to read more about this topic.

> Btw, https://www.understandinglatency.com today was great, hope more of you join us tomorrow. Jason will be giving an update on "L4S Field Trial Experiences", also Stuart "It's latency, stupid!" of Apple will be speaking and I'm looking forward to my compatriot, Marek Pesta, with CDN77, adding more knowledge to the CDN field.
> 
> All the best,
> 
> Frank
> 
> Frantisek (Frank) Borsik
> 
>  
> 
> https://www.linkedin.com/in/frantisekborsik
> 
> Signal, Telegram, WhatsApp: +421919416714 
> 
> iMessage, mobile: +420775230885
> 
> Skype: casioa5302ca
> 
> frantisek.borsik at gmail.com
> 
> 
> 
> On Mon, Dec 11, 2023 at 4:04 PM Sebastian Moeller <moeller0 at gmx.de> wrote:
> Hi Frantisek,
> 
> still keeping NNagain off-list, no need to infect yet another list with my L4S aversion ;)
> 
> > On Dec 11, 2023, at 11:18, Frantisek Borsik <frantisek.borsik at gmail.com> wrote:
> > 
> > …and I’m adding NN again, because:
> > 
> > No matter which evil end game is actually the one, it will create a quite a havoc.
> 
>         Not so sure about that, my expectancy is mostly that L4S is going to fizzle out simply because it over-promised and under-delivers. And fizzling out does not make impressive havoc...
> 
> > Seeing, for example Netflix, participating in L4S trail (they said it was for their cloud gaming,
> 
>         But if we look at the DualQ from the outside we see a priority scheduler with little to no organic bulk traffic (yet?) so using this for any latency sensitive traffic is tempting. The bigger question for me is, does Netflix' cloud gaming traffic actually respond gracefully to CE marks as expected for an L5S flow, or if it is using NQB instead is it following the eligibility recommendations for NQB traffic? That is, is this traffic playing along the rules or not? (One of the problems with L4S is that it defines a set of requirements for traffic and then fails to enforce/police these requirements, making them pretty irrelevant, but I digress).
> 
> 
> > while at IETF 118) is in my mind the same like seeing a capitalist, an entrepreneur selling a rope to a bloodthirsty communist that want to create a noose to hang him on it :-) but if one don’t see/hear evil…there is no hope.
> 
>         I see less diabolic evil in L4S and more ideologically* driven mediocrity...
> 
> Regards
>         Sebastian
> 
> 
> *) Hard to read https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=6e7364c67c13fa991c63292b281fac61dfcf40e0 "Flow Rate Fairness: Dismantling a Religion" as anything but ideologically driven...
> 
> > 
> > 
> > All the best,
> > 
> > Frank
> > Frantisek (Frank) Borsik
> > 
> > https://www.linkedin.com/in/frantisekborsik
> > Signal, Telegram, WhatsApp: +421919416714 
> > iMessage, mobile: +420775230885
> > Skype: casioa5302ca
> > frantisek.borsik at gmail.com
> > 
> > 
> > On Mon, 11 Dec 2023 at 10:52 AM, Sebastian Moeller <moeller0 at gmx.de> wrote:
> > Hi Frantisek,
> > 
> > steering this to the bloat list, as I have a hunch my response might not be appropriate for the wide audience in NNagain even though this can have NN ramifications.
> > 
> > 
> > > On Dec 11, 2023, at 10:50, Frantisek Borsik via Nnagain <nnagain at lists.bufferbloat.net> wrote:
> > > 
> > > Top story on HN:
> > > 
> > > h++ps://news.ycombinator.com/item?id=38597744
> > > 
> > > My pick for top comment:
> > > 
> > > <image_6487327.JPG>
> > > 
> > > 
> > > A possible L4S end game? 😁 
> > 
> > Unlikely, as that would e.g. violate EU directive 2015/2120 too blatantly. My take is more subtle, L4S is essentially a technique that only works well for short RTTs and the end game is to make datacenter space inside eyeball-ISPs much more desirable and hence yielding higher prices... L4S effectively increases TCPs existing RTT bias* and hence makes long range L4S connections undesirable/under-performing, and that implies that content providers that want to use L4S will need to get their servers/CDNs close to the eye-balls. But hey, I would love to be wrong and ISPs jus do this out of the goodness of their hearts... (only then I would politely ask the to deploy a flow queueing AQM instead of DualQ, as that in one swoop will solve a ship-load of L4S problems).
> > 
> > Regards
> >         Sebastian
> > 
> > 
> > *) L4S argues that this is not an increase, but that due to removing traditional TCP's excess queueing it just reveals TCPs inherent RTT bias; IMHO that is a bull shit argument since the root cause is not important, what matters is that with L4S RTT bias becomes measurably worse.
> > 
> > 
> > P.S.: I think my mail provider marks outgoing mails with too many links as SPAM, so I need to make the URLs invisible, and I believe replacing the t's in h t t p with plus signs does the trick... 
> > 
> > 
> > > 
> > > 
> > > All the best,
> > > 
> > > Frank
> > > Frantisek (Frank) Borsik
> > > 
> > > h++ps://www.linkedin.com/in/frantisekborsik
> > > Signal, Telegram, WhatsApp: +421919416714 
> > > iMessage, mobile: +420775230885
> > > Skype: casioa5302ca
> > > frantisek.borsik at gmail.com
> > > 
> > > 
> > > On Sun, 10 Dec 2023 at 7:22 PM, Robert McMahon via Nnagain <nnagain at lists.bufferbloat.net> wrote:
> > > Sorry, this isn't a very good technical analysis. It's a push agenda for AQM. I get that many people's egos are tied to their belief that my algorithm came down from God himself. It's not true though. 
> > > 
> > > 
> > > Bob
> > > On Dec 10, 2023, at 3:12 AM, Sauli Kiviranta via Nnagain <nnagain at lists.bufferbloat.net> wrote:
> > > I find it alarming for the future of internet that there is a trend to
> > > use elbowing techniques to squeeze metrics out at the expense of all
> > > other flows. We know that when nobody plays fair the arms race ends
> > > badly.
> > > 
> > > Best regards,
> > > Sauli
> > > 
> > > On 12/10/23, Frantisek Borsik via Nnagain <nnagain at lists.bufferbloat.net> wrote:
> > >  Well, to be completely honest, I would say its about general public, really
> > >  - it’s an article in The Verge :-)
> > > 
> > >  But yeah. Here’s something they should address:
> > > 
> > >  h++ps://github.com/heistp/l4s-tests#key-findings
> > > 
> > >  In general - a potential harm to innocent bystanders is a biggest L4S
> > >  problem:
> > >  h++ps://www.linkedin.com/posts/frantisekborsik_jonathanmorton-l4s-ietf118-activity-7128317873571176448-GWAX
> > > 
> > > 
> > > 
> > > 
> > >  All the best,
> > > 
> > >  Frank
> > >  Frantisek (Frank) Borsik
> > > 
> > >  h++ps://www.linkedin.com/in/frantisekborsik
> > >  Signal, Telegram, WhatsApp: +421919416714
> > >  iMessage, mobile: +420775230885
> > >  Skype: casioa5302ca
> > >  frantisek.borsik at gmail.com
> > > 
> > > 
> > >  On Sun, 10 Dec 2023 at 5:48 AM, Matt Mathis via Nnagain <
> > >  nnagain at lists.bufferbloat.net> wrote:
> > > 
> > >  About L4S
> > > 
> > >  h++ps://www.theverge.com/23655762/l4s-internet-apple-comcast-latency-speed-bandwidth
> > > 
> > >  It says something about their target audience that they feel they need to
> > >  explain bytes vs bits.
> > > 
> > >  Thanks,
> > >  --MM--
> > >  Evil is defined by mortals who think they know "The Truth" and use force
> > >  to apply it to others.
> > > 
> > >  Matt Mathis  (Email is best)
> > >  Home & mobile: 412-654-7529 please leave a message if you must call.
> > > 
> > > 
> > >  Nnagain mailing list
> > >  Nnagain at lists.bufferbloat.net
> > >  h++ps://lists.bufferbloat.net/listinfo/nnagain
> > > 
> > > 
> > > 
> > > Nnagain mailing list
> > > Nnagain at lists.bufferbloat.net
> > > h++ps://lists.bufferbloat.net/listinfo/nnagain
> > > _______________________________________________
> > > Nnagain mailing list
> > > Nnagain at lists.bufferbloat.net
> > > h++ps://lists.bufferbloat.net/listinfo/nnagain
> > > _______________________________________________
> > > Nnagain mailing list
> > > Nnagain at lists.bufferbloat.net
> > > h++ps://lists.bufferbloat.net/listinfo/nnagain
> > 
> 