From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from smtp121.iad3a.emailsrvr.com (smtp121.iad3a.emailsrvr.com [173.203.187.121]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by lists.bufferbloat.net (Postfix) with ESMTPS id 7D0093B29E for ; Sat, 4 Dec 2021 18:01:51 -0500 (EST) Received: from app36.wa-webapps.iad3a (relay-webapps.rsapps.net [172.27.255.140]) by smtp32.relay.iad3a.emailsrvr.com (SMTP Server) with ESMTP id 152F41963; Sat, 4 Dec 2021 18:01:51 -0500 (EST) Received: from deepplum.com (localhost.localdomain [127.0.0.1]) by app36.wa-webapps.iad3a (Postfix) with ESMTP id 0075D603D7; Sat, 4 Dec 2021 18:01:51 -0500 (EST) Received: by apps.rackspace.com (Authenticated sender: dpreed@deepplum.com, from: dpreed@deepplum.com) with HTTP; Sat, 4 Dec 2021 18:01:50 -0500 (EST) X-Auth-ID: dpreed@deepplum.com Date: Sat, 4 Dec 2021 18:01:50 -0500 (EST) From: "David P. Reed" To: "Jonathan Morton" Cc: "Dave Taht" , jonathan.kua@deakin.edu.au, "Cake List" MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="----=_20211204180151000000_16568" Importance: Normal X-Priority: 3 (Normal) X-Type: html In-Reply-To: <0A6AB0B7-E010-42E3-BAEE-FCBFA5995117@gmail.com> References: <0A6AB0B7-E010-42E3-BAEE-FCBFA5995117@gmail.com> X-Client-IP: 209.6.168.128 Message-ID: <1638658910.996417608@apps.rackspace.com> X-Mailer: webmail/19.0.13-RC X-Classification-ID: 02c8a524-ca49-4dda-9405-a04da427001a-1-1 Subject: Re: [Cake] Understanding the Achieved Rate Multiplication Effect in FlowQueue-based AQM Bottleneck X-BeenThere: cake@lists.bufferbloat.net X-Mailman-Version: 2.1.20 Precedence: list List-Id: Cake - FQ_codel the next generation List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sat, 04 Dec 2021 23:01:51 -0000 ------=_20211204180151000000_16568 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable =0AI agree with your broad assessment, Jonathan.=0A =0AThe self-interferenc= e problem within a host isn't just a network problem. It's a user-space sch= eduler problem as well.=0A =0AThere are lots of interactions between user-s= pace scheduler (in the case of Linux, the "Completely Fair Scheduler" and i= ts quantum, which is set by the HZ variable at boot) and the network stack = in the kernel. This interactions have non-trivial effects when mutliple flo= ws are independently created by concurrent processes).=0A =0ALately, I've b= een studying, for reasons related to my day job, the complex interactions o= f timing at sub-millisecond scale among threads and processes on a single s= ystem in Linux. I/O driven by threads become highly correlated, and so assu= ming "independence" among flow timing is just not a good assumption.=0A = =0AThe paper observes the results of "dependencies" that couple/resonate.= =0A =0AOn Friday, December 3, 2021 7:09pm, "Jonathan Morton" said:=0A=0A=0A=0A> > On 4 Dec, 2021, at 12:27 am, Dave Taht wrote:=0A> >=0A> >=0A> https://jonathankua.github.io/prepr= ints/jkua-ieeelcn2021_understanding_ar_preprint-20jul2021.pdf=0A> >=0A> > I= would love it if somehow the measured effects of chunklets against cake's= =0A> per-host/per flow fq was examined one day.=0A> =0A> I haven't actually= measured it, but based on what the above paper says, I can make=0A> some f= irm predictions:=0A> =0A> 1: When competing against traffic to the same loc= al host, the performance effects=0A> they describe will be present.=0A> =0A= > 2: When competing against traffic to a different local-network host, the= =0A> performance effects they describe will be attenuated or even entirely = absent.=0A> =0A> 3: They noted one or two cases of observable effects of ha= sh collisions in their=0A> tests with FQ-Codel. These will be greatly reduc= ed in prevalence with Cake, due=0A> to the set-associative hash function wh= ich specifically addresses that phenomenon.=0A> =0A> - Jonathan Morton=0A> = _______________________________________________=0A> Cake mailing list=0A> C= ake@lists.bufferbloat.net=0A> https://lists.bufferbloat.net/listinfo/cake= =0A> ------=_20211204180151000000_16568 Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable

I agree with your broa= d assessment, Jonathan.

=0A

 

=0A

The self-interference problem within a host isn't just a network = problem. It's a user-space scheduler problem as well.

=0A

 

=0A

There are lots of interactions betw= een user-space scheduler (in the case of Linux, the "Completely Fair Schedu= ler" and its quantum, which is set by the HZ variable at boot) and the netw= ork stack in the kernel. This interactions have non-trivial effects wh= en mutliple flows are independently created by concurrent processes).

= =0A

 

=0A

Lately, I've been= studying, for reasons related to my day job, the complex interactions of t= iming at sub-millisecond scale among threads and processes on a single syst= em in Linux. I/O driven by threads become highly correlated, and so assumin= g "independence" among flow timing  is just not a good assumption.

= =0A

 

=0A

The paper observe= s the results of "dependencies" that couple/resonate.

=0A

 

=0A

On Friday, December 3, 2021 7:09pm,= "Jonathan Morton" <chromatix99@gmail.com> said:

=0A=0A

> > On 4 Dec, = 2021, at 12:27 am, Dave Taht <dave.taht@gmail.com> wrote:
> &= gt;
> >
> https://jonathankua.github.io/preprints/jkua-i= eeelcn2021_understanding_ar_preprint-20jul2021.pdf
> >
>= > I would love it if somehow the measured effects of chunklets against = cake's
> per-host/per flow fq was examined one day.
>
> I haven't actually measured it, but based on what the above paper say= s, I can make
> some firm predictions:
>
> 1: When= competing against traffic to the same local host, the performance effects<= br />> they describe will be present.
>
> 2: When compe= ting against traffic to a different local-network host, the
> perfo= rmance effects they describe will be attenuated or even entirely absent.>
> 3: They noted one or two cases of observable effects of = hash collisions in their
> tests with FQ-Codel. These will be great= ly reduced in prevalence with Cake, due
> to the set-associative ha= sh function which specifically addresses that phenomenon.
>
&= gt; - Jonathan Morton
> ___________________________________________= ____
> Cake mailing list
> Cake@lists.bufferbloat.net
= > https://lists.bufferbloat.net/listinfo/cake
>

=0A ------=_20211204180151000000_16568--