From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-wi0-f175.google.com (mail-wi0-f175.google.com [209.85.212.175]) (using TLSv1 with cipher RC4-SHA (128/128 bits)) (Client did not present a certificate) by huchra.bufferbloat.net (Postfix) with ESMTPS id 16F2C21F0F4 for ; Wed, 5 Sep 2012 20:31:55 -0700 (PDT) Received: by wibhm2 with SMTP id hm2so4369421wib.10 for ; Wed, 05 Sep 2012 20:31:54 -0700 (PDT) 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:content-transfer-encoding; bh=sqro71A+m43GupExA2vjp/1denlgdE3J4TaX/p3FIIA=; b=q3qE3tNBfVqH3a8j2Y6QrqC4lI/4TnAD7AVjTJIm4yV5bPPQrHHC/h2euV6q/IlKRX eA3iOlEMnVRMN2ZVmzdM9Pgs5Zh/OTGiN4MV3pTh/SDPNF0YQU2ZaDVjhqs1UQvuRyoO k/mE4PYXcxj2L4N+nBETRTZhVQpnaBjCKZSjqZy3iOA8/jnNtQpjOEZK5BHQRv6DSSXM Al0aiMFptvi9ISJREn/KEOW8lRl71uISZCIgNCJNSRvDNnavvO5yV2VZYY1HZzdVCoFZ nRwO1nADZTx7ZZB5blp7kXuoBbQRGVwvmd3dTSJliNYLFEseN79wF0G43PiCC87+3jpG 7CQA== MIME-Version: 1.0 Received: by 10.180.74.33 with SMTP id q1mr42575107wiv.4.1346902314172; Wed, 05 Sep 2012 20:31:54 -0700 (PDT) Received: by 10.223.159.134 with HTTP; Wed, 5 Sep 2012 20:31:54 -0700 (PDT) In-Reply-To: References: Date: Wed, 5 Sep 2012 20:31:54 -0700 Message-ID: From: Dave Taht To: david@lang.hm, jow@openwrt.org Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable Cc: cerowrt-devel@lists.bufferbloat.net Subject: Re: [Cerowrt-devel] what's the status of pushing things upstream X-BeenThere: cerowrt-devel@lists.bufferbloat.net X-Mailman-Version: 2.1.13 Precedence: list List-Id: Development issues regarding the cerowrt test router project List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 06 Sep 2012 03:31:56 -0000 On Wed, Sep 5, 2012 at 1:46 PM, wrote: > I'm seeing discussion on the openwrt mailing list that they are goin to > actually make a release ;) I've been too heads-down trying to fix stuff to track the state of the free= ze. I have a test deployment going of openwrt and cerowrt that I started last month, on nanostation M5, picostation 2HP, and wndr3700v2 hardware, at a local campground, with real users. It's working well except the 32MB devices are running into some memory and OOM issues, which I hopefully have fixed at this point, except that some of the fixes are rather hacky. The most effective one thus far seems to be the skb_reduce_truesize patch in the qdisc. Std openwrt won't run into the memory problems as rapidly as I did with all the extra stuff on by default in cero.... I'd hoped to do a new test deployment this past week, but some persistent (and non-technical) issues beat me. > how much of the cerowrt code changes are upstream and will be in this > release vs stuff that isn't ready or isn't going to make it? 1) What's in openwrt right now could use a little work to avoid surprises. Specifically: 1a) Qos-scripts should put a packet limit on fq_codel. 1200 or so, works for me on rates below 30Mbit. For upload rates below 100Mbit at this point, I'd recommend turning ecn off, too. It makes no sense to turn ecn off on downloads. I've also been tuning the quantum down. That's all I care make it upstream pre-freeze at this juncture. Other stuff is all upstream and working well: codel and fq_codel are in openwrt's default kernel. fq_codel is used in qos-scripts. fq_codel is not on by default elsewhere in openwrt. Tons of ipv6, classification, etc fixes through the stack. The one that I think holds the most promise is the addition of AAAA record insertion and dhcpv6 support to dnsmasq, but overall, it's tons better than last year's stuff. Stuff that will need work after the freeze: 0) BQL support is available in only some of the device drivers openwrt is shipping (not enough) I would really like a big BQL push on key ethernet drivers - things like the dreamplug and the raspberry pi don't have it yet. I don't know what other devices are popular.... It's just a couple lines of code that make a HUGE difference. 1) I think it's possible to do much better than qos-scripts. 2) HTB and I are not having a good time together at higher rates on this hw= . 3) fq_codel and codel have some fixes pending in 3.6 and in my tree. (see codel list for some discussion). I have the ns2 model based codel which appears to be better in most respects than the current one, and some thoughts towards some fq_codel variants. A great deal of benchmarking and testing remains to be done before that can go mainline. I do plan to release a cornucopia of fq_codel variants in the next cerowrt snapshot. 4) cerowrt runs fq_codel on all devices and all queues by default. Openwrt doesn't. I've been trying to get to where what I had was stable enough to push up there before they released, but I am not quite there yet. It seems unlikely I'll get there this month, either. So just patching up qos-scripts would be good, for now. But basically you can add the debloat package to any openwrt hw (with adaquate memory!), and IF the ethernet device driver has BQL support, you'll see an improvement. I think the ADSL fixes from david woodhouse made this release of openwrt, too, so debloat will help there, but haven't checked. On wifi, the situation is far more nuanced, and it's pretty obvious we have a long way to go there. I figure we're in for another year of promise on top of the pain. Minimum. But I'm seriously encouraged. There is a hint of dim lights at the end of the tunnel. > David Lang > _______________________________________________ > Cerowrt-devel mailing list > Cerowrt-devel@lists.bufferbloat.net > https://lists.bufferbloat.net/listinfo/cerowrt-devel --=20 Dave T=E4ht http://www.bufferbloat.net/projects/cerowrt/wiki - "3.3.8-17 is out with fq_codel!"