From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from full.lackof.org (full.lackof.org [204.13.164.203]) (using TLSv1.2 with cipher ADH-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by lists.bufferbloat.net (Postfix) with ESMTPS id 4F8ED3B2A4 for ; Thu, 2 Jun 2022 16:29:14 -0400 (EDT) Received: from [172.16.1.4] (97-126-25-158.tukw.qwest.net [97.126.25.158]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (2048 bits) server-digest SHA256) (Client did not present a certificate) by full.lackof.org (Postfix) with ESMTPSA id 4LDd0W20rPzyrW; Thu, 2 Jun 2022 20:29:31 +0000 (UTC) Message-ID: <9aedf6d7-d5f2-c410-17fa-da01a0dae380@lackof.org> Date: Thu, 2 Jun 2022 13:29:12 -0700 MIME-Version: 1.0 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101 Thunderbird/91.8.0 Content-Language: en-US To: cerowrt-devel@lists.bufferbloat.net References: <1654189891.625331931@apps.rackspace.com> From: Matt Taggart In-Reply-To: <1654189891.625331931@apps.rackspace.com> Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 8bit Subject: Re: [Cerowrt-devel] 2.5gbit for $59 X-BeenThere: cerowrt-devel@lists.bufferbloat.net X-Mailman-Version: 2.1.20 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, 02 Jun 2022 20:29:14 -0000 On 6/2/22 10:11, David P. Reed wrote: > There are small, low-TDP Intel systems for  up to ~$250 or so (including > case) that use current generation Celerons with 4 2.5 GigE ports, and > with the I/O bandwidth to easily support a full-on router at wirespeed > on those ports. > > I'm thinking of upgrading my entry-router (which is based on Fedora > Server 36 now, not Cerowrt, just because that's my general go-to distro > on x86_64 and Aarch64) from an old Celeron system with two full speed 1 > GigE ports to 2.5 GigE, in advance of my expectation that 2.5 GigE > DOCSIS 3.1 will become cheap enough soon at my home. > > The problem with the low-end boards is that you need enough PCIe lanes > to move packets at 10 Gb/sec bidirectionally. The contained ARM chips > may be fast enough in principle, but the board and the PCIe are a > bottleneck. > > AliExpress sells such boards and also barebones, but prices and specs vary. The ones I see there seem to be using Celeron N5090 or N5105. Both have "PCI Express 3.0 controller supporting 8 lanes (multiplexed); 4 lanes available externally" They all seem to be using "4x Intel i225-V" Apparently earlier revs of that had problems but the "B3" stepping is supposed to be fixed. Each uses pci-e 3.1 x1. So depending on how the board is laid out, they should have the bandwidth to actually do 2.5Gbit. All the usb ports, wifi, graphics, etc should all be using the internal lanes I think. Here is a comparison of those celerons, the nanopi, and the pi4 https://www.cpubenchmark.net/compare/ROC-RK3568-PC-HDMI-(Android)-vs-BCM2711-vs-Intel-Celeron-N5105-vs-Intel-Celeron-N5095/4752vs4297vs4412vs4472 more details on the specific pages. The nanopi seems mostly better than the pi4, except some floating point and matrix. The Celerons are much better CPUs, but are in a different power consumption and price class. -- Matt Taggart matt@lackof.org