From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-vc0-x229.google.com (mail-vc0-x229.google.com [IPv6:2607:f8b0:400c:c03::229]) (using TLSv1 with cipher RC4-SHA (128/128 bits)) (Client CN "smtp.gmail.com", Issuer "Google Internet Authority G2" (verified OK)) by huchra.bufferbloat.net (Postfix) with ESMTPS id A8B7C21F322 for ; Tue, 3 Mar 2015 11:34:02 -0800 (PST) Received: by mail-vc0-f169.google.com with SMTP id im6so6340304vcb.0 for ; Tue, 03 Mar 2015 11:34:01 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type; bh=w4WCbX9FmRsguMrGfH3HXgNq8kiRB2yhPRLBUpoxtbk=; b=yU8NBYWCXiF2x14X9z2irEYaQjDLRUOCYA/7N8Yjqq0/5hDOsP/nZOfLULMbt8+AfT hGmbEodmWz2OtGnetmHPSvYbKX6j19f/Ju2Eqj8/MlwQarmfOv+Tq5TFfWlb/yzEIG7E rX9s1ZJz1cY9HnmnqD7WagNYuUh3wkjMwJZo+DjpPT0Glrzutd3J3rwHLq9fGkw+nY1y 6V3UelB/nUZHavFRo0UPlKUY3eifbisVg2cqQ3Q2kfIwtW4IhP0L9G8ez/cUZ8U7ydso 3/CSYxdj8N9sioWIiBAI/8l1ga7dK7UCMSeI6CObzhPxNeWQk7cpgYd6p2SYJhs3IQc1 a/0g== MIME-Version: 1.0 X-Received: by 10.52.243.41 with SMTP id wv9mr724777vdc.20.1425411241304; Tue, 03 Mar 2015 11:34:01 -0800 (PST) Received: by 10.52.24.79 with HTTP; Tue, 3 Mar 2015 11:34:01 -0800 (PST) Received: by 10.52.24.79 with HTTP; Tue, 3 Mar 2015 11:34:01 -0800 (PST) In-Reply-To: References: Date: Tue, 3 Mar 2015 21:34:01 +0200 Message-ID: From: Jonathan Morton To: mandy ahuja Content-Type: multipart/alternative; boundary=001a11340138b2694c0510676aea Cc: codel@lists.bufferbloat.net Subject: Re: [Codel] Regarding BDP, window size and queue-size for codel in ns2 X-BeenThere: codel@lists.bufferbloat.net X-Mailman-Version: 2.1.13 Precedence: list List-Id: CoDel AQM discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 03 Mar 2015 19:34:30 -0000 --001a11340138b2694c0510676aea Content-Type: text/plain; charset=UTF-8 I recommended earlier to run Codel experiments with a very large buffer, since Codel will keep it under control. For experimental rigour, you should also run experiments involving other queue modes with the same buffer size, and observe how well or badly each mode controls it. As for receive window size, you should make that large enough that you will notice the resulting delay easily. Modern operating systems will happily use receive windows of several megabytes in default configurations; this results in impressive delays at relatively low bandwidths. I've seen 45 seconds at 512Kbps. You might also try setting up multiple simulated hosts with different receive window sizes - including some sensibly small ones, such as the 65535 bytes which is the limit for TCP stacks without window scaling - and see how they compete on the same congested link. This might give you some insight into why such large windows seem desirable to those unaware of AQM, and why they become mostly irrelevant when good AQM is introduced. - Jonathan Morton --001a11340138b2694c0510676aea Content-Type: text/html; charset=UTF-8 Content-Transfer-Encoding: quoted-printable

I recommended earlier to run Codel experiments with a very l= arge buffer, since Codel will keep it under control. For experimental rigou= r, you should also run experiments involving other queue modes with the sam= e buffer size, and observe how well or badly each mode controls it.

As for receive window size, you should make that large enoug= h that you will notice the resulting delay easily. Modern operating systems= will happily use receive windows of several megabytes in default configura= tions; this results in impressive delays at relatively low bandwidths. I= 9;ve seen 45 seconds at 512Kbps.

You might also try setting up multiple simulated hosts with = different receive window sizes - including some sensibly small ones, such a= s the 65535 bytes which is the limit for TCP stacks without window scaling = - and see how they compete on the same congested link. This might give you = some insight into why such large windows seem desirable to those unaware of= AQM, and why they become mostly irrelevant when good AQM is introduced.

- Jonathan Morton

--001a11340138b2694c0510676aea--