[Bloat] The hell has frozen: Latency mentioned at APNIC 62, during a keynote!

[Bloat] The hell has frozen: Latency mentioned at APNIC 62, during a keynote!

[Frantisek Borsik]

"The scarce resource is not bandwidth, or compute. It is latency."


https://conference.apnic.net/62/assets/presentation-files/93297823-389f-4ff6-ba5f-73cfe12a367b.pdf

Incredible. Thanks to Mohit for catching this gem.

All the best,

Frank

Frantisek (Frank) Borsik


*In loving memory of Dave Täht: *1965-2025

https://libreqos.io/2025/04/01/in-loving-memory-of-dave/


https://www.linkedin.com/in/frantisekborsik

Signal, Telegram, WhatsApp: +421919416714

iMessage, mobile: +420775230885

Skype: casioa5302ca

frantisek.borsik@gmail.com

[Bloat] Re: The hell has frozen: Latency mentioned at APNIC 62, during a keynote!

[Stuart Cheshire]

> On Sep 10, 2026, at 10:37, Frantisek Borsik <frantisek.borsik@gmail.com> wrote:
> 
> "The scarce resource is not bandwidth, or compute. It is latency."
> 
> https://conference.apnic.net/62/assets/presentation-files/93297823-389f-4ff6-ba5f-73cfe12a367b.pdf
> 
> Incredible. Thanks to Mohit for catching this gem.

Nice, but I would not express it that way.

Calling latency a “scarce resource” sends the wrong message. It reinforces the incorrect idea that low-latency service is a rare commodity, in short supply, that can only be provided to the small privileged minority that can afford it. This is the wrong-headed thinking behind priorities, traffic classes, DiffServ, etc. People assume latency, like bandwidth, is a zero-sum game. (“Better latency for me means worse latency for you.”) The great insight of L4S is that latency is not a zero-sum game. *All* traffic can receive low latency, and that benefit is not counterbalanced by some corresponding degradation elsewhere. L4S with short queues is just universally better than tail-drop with long queues, and there is no hidden caveat or disclaimer on that statement.

Instead of the word “scarce” I would use “important” or “valuable”.

It is true that milliseconds are precious -- once wasted you can’t get them back, but there is no scarcity of them.

Stuart Cheshire

[Bloat] Re: The hell has frozen: Latency mentioned at APNIC 62, during a keynote!

[Neal Cardwell]

On Thu, Sep 17, 2026 at 11:24 AM Stuart Cheshire via Bloat <
bloat@lists.bufferbloat.net> wrote:

> > On Sep 10, 2026, at 10:37, Frantisek Borsik <frantisek.borsik@gmail.com>
> wrote:
> >
> > "The scarce resource is not bandwidth, or compute. It is latency."
> >
> >
> https://conference.apnic.net/62/assets/presentation-files/93297823-389f-4ff6-ba5f-73cfe12a367b.pdf
> >
> > Incredible. Thanks to Mohit for catching this gem.
>
> Nice, but I would not express it that way.
>
> Calling latency a “scarce resource” sends the wrong message. It reinforces
> the incorrect idea that low-latency service is a rare commodity, in short
> supply, that can only be provided to the small privileged minority that can
> afford it. This is the wrong-headed thinking behind priorities, traffic
> classes, DiffServ, etc. People assume latency, like bandwidth, is a
> zero-sum game. (“Better latency for me means worse latency for you.”) The
> great insight of L4S is that latency is not a zero-sum game. *All* traffic
> can receive low latency, and that benefit is not counterbalanced by some
> corresponding degradation elsewhere. L4S with short queues is just
> universally better than tail-drop with long queues, and there is no hidden
> caveat or disclaimer on that statement.
>
> Instead of the word “scarce” I would use “important” or “valuable”.
>
> It is true that milliseconds are precious -- once wasted you can’t get
> them back, but there is no scarcity of them.


Well said.

Likewise, I would not call latency a "resource". The link is the resource.
Latency is a performance characteristic. Effective management of the
resource (with L4S or other schemes) can yield both low latency and high
throughput.

Calling latency a "resource" makes it sound like something that can be
built or purchased. One cannot build or purchase low latency. You can
purchase hardware that allows low latency, but you still have to deploy
protocols that know how to use the signals to achieve low latency and high
throughput.

So rather than saying: "The scarce resource is not bandwidth, or compute.
It is latency." ... perhaps: "The most important performance metric where
systems fall short today is not bandwidth, or compute. It is latency."

neal

[Bloat] Re: The hell has frozen: Latency mentioned at APNIC 62, during a keynote!

[Richard Scheffenegger]

In other news, ultra ethernet is using credit based flow control (to push queues back into the sender at much higher impact on efficiency if fibrechannel is any guidance)

Best regards,
   Richard Scheffenegger

> Am 17.09.2026 um 21:57 schrieb Neal Cardwell via Bloat <bloat@lists.bufferbloat.net>:
> 
> On Thu, Sep 17, 2026 at 11:24 AM Stuart Cheshire via Bloat <
> bloat@lists.bufferbloat.net> wrote:
> 
>>> On Sep 10, 2026, at 10:37, Frantisek Borsik <frantisek.borsik@gmail.com>
>> wrote:
>>> 
>>> "The scarce resource is not bandwidth, or compute. It is latency."
>>> 
>>> 
>> https://conference.apnic.net/62/assets/presentation-files/93297823-389f-4ff6-ba5f-73cfe12a367b.pdf
>>> 
>>> Incredible. Thanks to Mohit for catching this gem.
>> 
>> Nice, but I would not express it that way.
>> 
>> Calling latency a “scarce resource” sends the wrong message. It reinforces
>> the incorrect idea that low-latency service is a rare commodity, in short
>> supply, that can only be provided to the small privileged minority that can
>> afford it. This is the wrong-headed thinking behind priorities, traffic
>> classes, DiffServ, etc. People assume latency, like bandwidth, is a
>> zero-sum game. (“Better latency for me means worse latency for you.”) The
>> great insight of L4S is that latency is not a zero-sum game. *All* traffic
>> can receive low latency, and that benefit is not counterbalanced by some
>> corresponding degradation elsewhere. L4S with short queues is just
>> universally better than tail-drop with long queues, and there is no hidden
>> caveat or disclaimer on that statement.
>> 
>> Instead of the word “scarce” I would use “important” or “valuable”.
>> 
>> It is true that milliseconds are precious -- once wasted you can’t get
>> them back, but there is no scarcity of them.
> 
> 
> Well said.
> 
> Likewise, I would not call latency a "resource". The link is the resource.
> Latency is a performance characteristic. Effective management of the
> resource (with L4S or other schemes) can yield both low latency and high
> throughput.
> 
> Calling latency a "resource" makes it sound like something that can be
> built or purchased. One cannot build or purchase low latency. You can
> purchase hardware that allows low latency, but you still have to deploy
> protocols that know how to use the signals to achieve low latency and high
> throughput.
> 
> So rather than saying: "The scarce resource is not bandwidth, or compute.
> It is latency." ... perhaps: "The most important performance metric where
> systems fall short today is not bandwidth, or compute.

[Bloat] Re: The hell has frozen: Latency mentioned at APNIC 62, during a keynote!

[Stuart Cheshire]

On Sep 17, 2026, at 12:56, Neal Cardwell <ncardwell@google.com> wrote:

> Likewise, I would not call latency a "resource". The link is the resource. Latency is a performance characteristic. Effective management of the resource (with L4S or other schemes) can yield both low latency and high throughput.
> 
> Calling latency a "resource" makes it sound like something that can be built or purchased. One cannot build or purchase low latency. You can purchase hardware that allows low latency, but you still have to deploy protocols that know how to use the signals to achieve low latency and high throughput.

Very good point. This is 100% right.

A “resource” is something that exists in finite quantity, so you need to decide where to allocate that limited resource. Almost by definition, I think calling something a “resource” is synonymous with saying that its allocation is a zero-sum game.

Stuart Cheshire