Speed tests try to determine the characteristics of the last mile connection, using some assumption about what happens when you reach its capacity, like increased queuing delay or increased packet loss. They probably work better when there is a physical limit like DSL line rate. Speed tiers can be problematic, since there is often at least some burst above the subscribed rate. I see this particularly on the speedtest.net upstream graph, which will jump up to a high number and then gradually drop the test rate until it finds the sustained rate, but customers interpret the graph to mean the “speed” is not consistent.
Speed tests also don’t account for typical use which is lots of different content sharing the connection. Just because your 100M connection doesn’t pull 100M from the speedtest.net server doesn’t necessarily mean you can’t achieve 100M of typical mixed traffic. At least this new test is clear about what it is testing: how much traffic can you pull from Netflix. And given how Netflix dominates Internet traffic, that’s not an unrealistic thing to test. From: Daniel White Sent: Thursday, May 19, 2016 7:46 AM To: af@afmug.com Subject: Re: [AFMUG] fast.com utility So those Comcast results were from last night. Just ran again this morning… Daniel White Managing Director – Hardware Distribution Sales ConVergence Technologies Cell: +1 (303) 746-3590 dwh...@converge-tech.com From: Daniel White [mailto:afmu...@gmail.com] Sent: Thursday, May 19, 2016 6:42 AM To: 'af@afmug.com' <af@afmug.com> Subject: RE: [AFMUG] fast.com utility On my Comcast connection its within a few Mbps. 100Mbps on Fast.com and 105Mbps on Speedtest. Need to try my WISP connection later today. Daniel White Managing Director – Hardware Distribution Sales ConVergence Technologies Cell: +1 (303) 746-3590 dwh...@converge-tech.com From: Af [mailto:af-boun...@afmug.com] On Behalf Of CBB - Jay Fuller Sent: Wednesday, May 18, 2016 5:09 PM To: af@afmug.com Subject: Re: [AFMUG] fast.com utility Judging from the last 5 posts, no one has yet... if his test server is inhouse as i think he said - i agree, should be an interesting response ----- Original Message ----- From: Nate Burke To: af@afmug.com Sent: Wednesday, May 18, 2016 5:52 PM Subject: Re: [AFMUG] fast.com utility So then do you get full line speed results from the test? On 5/18/2016 5:51 PM, Cassidy B. Larson wrote: Interesting though, from watching tcpdump while doing a speedtest, I’m seeing it hit BOTH of our local on-net Netflix appliances (over IPv6). On May 18, 2016, at 4:49 PM, Cassidy B. Larson <c...@infowest.com> wrote: 2620:108:700f::36f4:7ea4 and 205.251.244.235 are both Amazon IPs. Netflix uses a lot of EC2 stuff, so you’re not necessarily hitting their “cache” when you pull up their website. On May 18, 2016, at 4:45 PM, Sterling Jacobson <sterl...@avative.net> wrote: Very inaccurate too. I get 160Mbps results on a 10Gbps connection. This is with a path to Netflix that pretty much sits in LA. So I am assuming I hit their CDN in LA all the time. Not sure where their speed test web app is located. IPv6 C:\Users\Sterling>tracert netflix.com Tracing route to netflix.com [2620:108:700f::36f4:7ea4] over a maximum of 30 hops: 1 <1 ms <1 ms <1 ms 2606:cb80:2:2::1 2 <1 ms <1 ms <1 ms 2604:ba00:1:22::1 3 18 ms 22 ms 22 ms he.net.slix.net [2607:fa18:1:f00::15] 4 18 ms 18 ms 19 ms 10ge1-1.core1.las1.he.net [2001:470:0:27d::1] 5 23 ms 23 ms 24 ms 10ge1-14.core1.lax2.he.net [2001:470:0:27e::1] 6 18 ms 21 ms 24 ms 100ge2-1.core1.lax1.he.net [2001:470:0:72::1] 7 16 ms 16 ms 16 ms asn-qwest-us-as209.10gigabitethernet5-5.core1.lax1.he.net [2001:470:0:2c0::2] 8 26 ms 26 ms 26 ms 2001:428::205:171:3:199 9 23 ms 24 ms 25 ms 2001:428:7000:10:0:16:0:2 10 * * * Request timed out. 11 42 ms 42 ms 43 ms 2620:107:3000::e 12 42 ms 42 ms 43 ms 2620:108:7000::6 13 42 ms 42 ms 43 ms 2620:108:7000::7 14 42 ms 42 ms 42 ms 2620:108:7000::1 15 * * * Request timed out. 16 * * * Request timed out. 17 * * * Request timed out. 18 * 42 ms 42 ms 2620:108:700f::36f4:7ea4 IPv4 C:\Users\Sterling>tracert -4 netflix.com Tracing route to netflix.com [54.225.192.83] over a maximum of 30 hops: 1 <1 ms <1 ms <1 ms 108-165-31-1.avative.net [108.165.31.1] 2 <1 ms <1 ms <1 ms tg1-8--200.br01.lsan.acedc.net [69.27.173.37] 3 4 ms 4 ms 5 ms 208.186.235.162 4 33 ms 33 ms 34 ms be-1.br02.chcgildt.integra.net [209.63.82.186] 5 32 ms 32 ms 32 ms equinix01-chi2.amazon.com [206.223.119.98] 6 38 ms 42 ms 42 ms 52.95.62.36 7 32 ms 32 ms 32 ms 52.95.62.49 8 52 ms 52 ms 52 ms 54.239.42.63 9 52 ms 52 ms 52 ms 54.239.42.69 10 * * * Request timed out. 11 * * * Request timed out. 12 54 ms 61 ms 67 ms 54.239.110.249 13 53 ms 53 ms 53 ms 54.239.111.105 14 53 ms 58 ms 55 ms 205.251.244.235 From: Af [mailto:af-boun...@afmug.com] On Behalf Of Eric Kuhnke Sent: Wednesday, May 18, 2016 4:33 PM To: af@afmug.com Subject: Re: [AFMUG] fast.com utility further discussion: https://news.ycombinator.com/item?id=11722775 This could be useful from a residential last mile customer point of view, to expose ISPs which have good peering/low congestion tospeedtest.net but might have less than optimal routing to Netflix. Or an ISP that is flat topping the traffic charts on an N x 10GbE link to netflix somewhere in the intermediate path. Some people will see radically different results from speetest vs this new Netflix test during peak evening hours. On Wed, May 18, 2016 at 3:24 PM, Nate Burke <n...@blastcomm.com> wrote: Just came across this https://fast.com. Utility from netflix. Torching it looks like it opens 3 HTTPS connections to 3 different IP Addresses to run the test. Only reports download speed, no Latency or Upload. Virus-free. www.avast.com
