Discussion of explicit congestion notification's impact on the Internet
 help / color / mirror / Atom feed
* [Ecn-sane] comments: "Enabling ECN for Datacenter Networks with RTT Variations"
@ 2019-12-30  4:02 Dave Taht
  2019-12-31  1:56 ` [Ecn-sane] [EXTERNAL] " Wei Bai
  2020-01-04 22:01 ` [Ecn-sane] " Dave Taht
  0 siblings, 2 replies; 3+ messages in thread
From: Dave Taht @ 2019-12-30  4:02 UTC (permalink / raw)
  To: ECN-Sane; +Cc: kaichen, webai

this ecn# & codel in p4 paper had probably of most interest to me four points:

1) Codel like algorithm on a tofino switch, implemented in p4
2) with some source code published... Not the p4 code, darn it...
unless I missed it?
   but a detailed description of how they used the timestamp field was neat
3 Honestly I'd never thought about how much rtt variation a dc had,
nor do I know if this model is correct
3) Not really clear what parameters the used for codel

https://resource.hpisys.com/seminar-papers/2019/paper/20191108_CoNEXT2019_Enabling%20ECN%20for%20Datacenter%20Networks%20with%20RTT%20Variations.pdf

"Despite being successful, prior ECN-based transports have an
important drawback: they adopt a fxed RTT value in calculating
instantaneous ECN marking threshold while overlooking the RTT
variations in practice. In this paper, we reveal that the current
practice of using a fxed high-percentile RTT for ECN threshold
calculation can lead to persistent queue buildups, signifcantly
increasing packet latency. On the other hand, directly adopting lower
percentile RTTs results in throughput degradation. To handle the
problem, we introduce ECN!, a simple yet efective solution to enable
ECN for RTT variations. At its heart, ECN! inherits the current
instantaneous ECN marking (based on a high-percentile RTT) to achieve
high throughput and burst tolerance, while further marking packets
(conservatively) upon detecting long-term queue buildups to eliminate
unnecessary queueing delay without degrading throughput. We implement
ECN! on a Barefoot Tofno switch"

-- 
Make Music, Not War

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-435-0729

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [Ecn-sane] [EXTERNAL] comments: "Enabling ECN for Datacenter Networks with RTT Variations"
  2019-12-30  4:02 [Ecn-sane] comments: "Enabling ECN for Datacenter Networks with RTT Variations" Dave Taht
@ 2019-12-31  1:56 ` Wei Bai
  2020-01-04 22:01 ` [Ecn-sane] " Dave Taht
  1 sibling, 0 replies; 3+ messages in thread
From: Wei Bai @ 2019-12-31  1:56 UTC (permalink / raw)
  To: Dave Taht, ECN-Sane; +Cc: kaichen, Junxue ZHANG

+Junxue 

-----Original Message-----
From: Dave Taht <dave.taht@gmail.com> 
Sent: Sunday, December 29, 2019 8:03 PM
To: ECN-Sane <ecn-sane@lists.bufferbloat.net>
Cc: kaichen@cse.ust.hk; Wei Bai <Wei.Bai@microsoft.com>
Subject: [EXTERNAL] comments: "Enabling ECN for Datacenter Networks with RTT Variations"

this ecn# & codel in p4 paper had probably of most interest to me four points:

1) Codel like algorithm on a tofino switch, implemented in p4
2) with some source code published... Not the p4 code, darn it...
unless I missed it?
   but a detailed description of how they used the timestamp field was neat
3 Honestly I'd never thought about how much rtt variation a dc had,
nor do I know if this model is correct
3) Not really clear what parameters the used for codel

https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fresource.hpisys.com%2Fseminar-papers%2F2019%2Fpaper%2F20191108_CoNEXT2019_Enabling%2520ECN%2520for%2520Datacenter%2520Networks%2520with%2520RTT%2520Variations.pdf&amp;data=02%7C01%7CWei.Bai%40microsoft.com%7C72909f0aab234fe9d7a308d78cdd2d4e%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637132754919827546&amp;sdata=xQpBmCOlUsWEDKL3UA3y4wMlzDj3eQw1O5a7Tb0gIrk%3D&amp;reserved=0

"Despite being successful, prior ECN-based transports have an
important drawback: they adopt a fxed RTT value in calculating
instantaneous ECN marking threshold while overlooking the RTT
variations in practice. In this paper, we reveal that the current
practice of using a fxed high-percentile RTT for ECN threshold
calculation can lead to persistent queue buildups, signifcantly
increasing packet latency. On the other hand, directly adopting lower
percentile RTTs results in throughput degradation. To handle the
problem, we introduce ECN!, a simple yet efective solution to enable
ECN for RTT variations. At its heart, ECN! inherits the current
instantaneous ECN marking (based on a high-percentile RTT) to achieve
high throughput and burst tolerance, while further marking packets
(conservatively) upon detecting long-term queue buildups to eliminate
unnecessary queueing delay without degrading throughput. We implement
ECN! on a Barefoot Tofno switch"

-- 
Make Music, Not War

Dave Täht
CTO, TekLibre, LLC
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.teklibre.com&amp;data=02%7C01%7CWei.Bai%40microsoft.com%7C72909f0aab234fe9d7a308d78cdd2d4e%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637132754919827546&amp;sdata=gB6EAR3riemq8XK02rsZYprZ0pnuUnk4RTNYDYQ%2FfM8%3D&amp;reserved=0
Tel: 1-831-435-0729

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [Ecn-sane] comments: "Enabling ECN for Datacenter Networks with RTT Variations"
  2019-12-30  4:02 [Ecn-sane] comments: "Enabling ECN for Datacenter Networks with RTT Variations" Dave Taht
  2019-12-31  1:56 ` [Ecn-sane] [EXTERNAL] " Wei Bai
@ 2020-01-04 22:01 ` Dave Taht
  1 sibling, 0 replies; 3+ messages in thread
From: Dave Taht @ 2020-01-04 22:01 UTC (permalink / raw)
  To: ECN-Sane; +Cc: kaichen, webai

On Sun, Dec 29, 2019 at 8:02 PM Dave Taht <dave.taht@gmail.com> wrote:
>
> this ecn# & codel in p4 paper had probably of most interest to me four points:
>
> 1) Codel like algorithm on a tofino switch, implemented in p4
> 2) with some source code published... Not the p4 code, darn it...
> unless I missed it?
>    but a detailed description of how they used the timestamp field was neat
> 3 Honestly I'd never thought about how much rtt variation a dc had,
> nor do I know if this model is correct

> 3) Not really clear what parameters the used for codel

For CoDel, the interval is set to 240µs while target is set to 10µs.

> https://resource.hpisys.com/seminar-papers/2019/paper/20191108_CoNEXT2019_Enabling%20ECN%20for%20Datacenter%20Networks%20with%20RTT%20Variations.pdf
>
> "Despite being successful, prior ECN-based transports have an
> important drawback: they adopt a fxed RTT value in calculating
> instantaneous ECN marking threshold while overlooking the RTT
> variations in practice. In this paper, we reveal that the current
> practice of using a fxed high-percentile RTT for ECN threshold
> calculation can lead to persistent queue buildups, signifcantly
> increasing packet latency. On the other hand, directly adopting lower
> percentile RTTs results in throughput degradation. To handle the
> problem, we introduce ECN!, a simple yet efective solution to enable
> ECN for RTT variations. At its heart, ECN! inherits the current
> instantaneous ECN marking (based on a high-percentile RTT) to achieve
> high throughput and burst tolerance, while further marking packets
> (conservatively) upon detecting long-term queue buildups to eliminate
> unnecessary queueing delay without degrading throughput. We implement
> ECN! on a Barefoot Tofno switch"
>
> --
> Make Music, Not War
>
> Dave Täht
> CTO, TekLibre, LLC
> http://www.teklibre.com
> Tel: 1-831-435-0729



-- 
Make Music, Not War

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-435-0729

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2020-01-04 22:01 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-12-30  4:02 [Ecn-sane] comments: "Enabling ECN for Datacenter Networks with RTT Variations" Dave Taht
2019-12-31  1:56 ` [Ecn-sane] [EXTERNAL] " Wei Bai
2020-01-04 22:01 ` [Ecn-sane] " Dave Taht

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox