From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from smtp69.iad3a.emailsrvr.com (smtp69.iad3a.emailsrvr.com [173.203.187.69]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by lists.bufferbloat.net (Postfix) with ESMTPS id D5B963B29D for ; Sat, 24 Jun 2023 14:41:52 -0400 (EDT) Received: from app34.wa-webapps.iad3a (relay-webapps.rsapps.net [172.27.255.140]) by smtp9.relay.iad3a.emailsrvr.com (SMTP Server) with ESMTP id 1BED4142A; Sat, 24 Jun 2023 14:41:52 -0400 (EDT) Received: from deepplum.com (localhost.localdomain [127.0.0.1]) by app34.wa-webapps.iad3a (Postfix) with ESMTP id 0494A620E0; Sat, 24 Jun 2023 14:41:52 -0400 (EDT) Received: by apps.rackspace.com (Authenticated sender: dpreed@deepplum.com, from: dpreed@deepplum.com) with HTTP; Sat, 24 Jun 2023 14:41:52 -0400 (EDT) X-Auth-ID: dpreed@deepplum.com Date: Sat, 24 Jun 2023 14:41:52 -0400 (EDT) From: "David P. Reed" To: "Cake List" Cc: "bloat" , "Dave Taht" MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="----=_20230624144152000000_85866" Importance: Normal X-Priority: 3 (Normal) X-Type: html X-Client-IP: 209.6.168.128 Message-ID: <1687632112.01524513@apps.rackspace.com> X-Mailer: webmail/19.0.24-RC X-Classification-ID: e3813393-869b-4ab8-b1b9-8d2e1a06f775-1-1 Subject: [Cake] Two questions re high speed congestion management and datagram protocols 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, 24 Jun 2023 18:41:52 -0000 ------=_20230624144152000000_85866 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable =0AI was recently looking at congestion control algorithms - endpoint-based= ones - that would deal with very low level, very low latency requirements = in datacenters that use high speed switch fabrics. (note: congestion contro= l in such datacenters is a very, very real issue, especially since some lay= er 2 switches are overbuffered by a huge factor, so they don't signal conge= stion while building very long queues at 40-400 Gb/sec)=0A =0AI came across= this work from MIT CSAIL [ https://ccp-project.github.io/ccp-guide/ ]( htt= ps://ccp-project.github.io/ccp-guide/ )=0A =0AI also was looking back to DC= CP as a useful way to get a UDP that handled congestion without engaging th= e higher layers, and preserving the other flexibility of UDP.=0A =0AAnyone = here have any experience with looking at the performance of these, especial= ly w.r.t. Cake, which operates at the IP layer and thus takes direct advant= age of IP congestion signalling? Does libreqos look like it might help?=0A = =0A[as the guy most associated with the creation of UDP, this remains an ar= ea of great interest personally, and is also presumably of relevance to QUI= C... I don't know if either of these things are supported in systems other = than Linux, but in server datacenters and my home lab, Linux networking is = all that matters]=0A =0A- David ------=_20230624144152000000_85866 Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable

I was recently looking= at congestion control algorithms - endpoint-based ones - that would deal w= ith very low level, very low latency requirements in datacenters that use h= igh speed switch fabrics. (note: congestion control in such datacenters is = a very, very real issue, especially since some layer 2 switches are overbuf= fered by a huge factor, so they don't signal congestion while building very= long queues at 40-400 Gb/sec)

=0A

 

=0A

I came across this work from MIT CSAIL https://ccp-project.github.io/ccp-gu= ide/

=0A

 

=0A

I als= o was looking back to DCCP as a useful way to get a UDP that handled conges= tion without engaging the higher layers, and preserving the other flexibili= ty of UDP.

=0A

 

=0A

Any= one here have any experience with looking at the performance of these, espe= cially w.r.t. Cake, which operates at the IP layer and thus takes direct ad= vantage of IP congestion signalling? Does libreqos look like it might help?=

=0A

 

=0A

[as the guy m= ost associated with the creation of UDP, this remains an area of great inte= rest personally, and is also presumably of relevance to QUIC... I don't kno= w if either of these things are supported in systems other than Linux, but = in server datacenters and my home lab, Linux networking is all that matters= ]

=0A

 

=0A

- David

<= /font> ------=_20230624144152000000_85866--