From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eddie Kohler Subject: Re: [RFC] dccp ccid-3: High-res or low-res timers? Date: Mon, 17 Nov 2008 11:27:29 -0800 Message-ID: <4921C5A1.5030609@cs.ucla.edu> References: <5bc4c4570810171021ua6371ebs1ffdf471382a8b13@mail.gmail.com> <20081105052733.GG6564@gerrit.erg.abdn.ac.uk> <5bc4c4570811060538h2d662507u5de1fb62c61cd569@mail.gmail.com> <20081106152048.GA3621@gerrit.erg.abdn.ac.uk> <20081106153824.GA9709@ghostprotocols.net> <5bc4c4570811060946j4d5d8d1mf88f8b92c72b59c7@mail.gmail.com> <5bc4c4570811061004nfc2afdcn6035d49ae654aef1@mail.gmail.com> <5bc4c4570811061017j3acef860vee84992e7295d06d@mail.gmail.com> <5bc4c4570811061405qe72e43cx861c537885804132@mail.gmail.com> <20081108085035.GA7112@gerrit.erg.abdn.ac.uk> <20081115105042.GA7798@gerrit.erg.abdn.ac.uk> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit To: Gerrit Renker , Leandro Sales , Arnaldo Carvalho de Melo , ian.mcdonald@jandi.co.nz, DCCP Mailing List Received: from out-66.smtp.ucla.edu ([169.232.46.167]:41182 "EHLO out-66.smtp.ucla.edu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751791AbYKQTfs (ORCPT ); Mon, 17 Nov 2008 14:35:48 -0500 In-Reply-To: <20081115105042.GA7798@gerrit.erg.abdn.ac.uk> Sender: netdev-owner@vger.kernel.org List-ID: Gerrit Renker wrote: > I would appreciate some advice and insights regarding the use of > high-resolution timers within a transport protocol, specifically > DCCP with CCID-3 (RFC 5348). > > ... > > Summing up, I have doubts that basing CCID-3 will bring advantages and > would much rather go the other way and (consistently) use lower resolution. > > Thoughts? I agree. If one way must be chosen, then choose lower resolution timers. The biggest potential problem with lower-resolution timers is that a sender's rate might be limited, not by network characteristics, but by timer resolution. But DCCP allows a fair amount of burstiness already. And there may be ways to avoid rate limitation in common cases without resorting to hrtimers. For example, a sending application could use a mixture of non-blocking system calls, allowing the sending application to "poke" the DCCP implementation on every scheduling. At any rate, it seems worth trying. Eddie