Forsi Posted April 14, 2023 CID Share Posted April 14, 2023 When testing 1gbps up/down fiber connection over wired ethernet via the automatic test on both a laptop and desktop I see bandwidth much lower than I'd typically see when using my own remote server or other speed test utilities. After clicking the 'test again' button 3-4 times the download test seems to cap out at around 600mbps, and the upload test seems capped at around 400mbps. I don't think I've ever seen the expected results on the first run. It seems as though this is skewing test history to show an average of much less bandwidth than I have been able to confirm having through other testing methodshttps://www.waveform.com/tools/bufferbloat?test-id=d6c40a41-56e8-4770-ba94-756ae659094d Why do you need to test a connection that has that much bandwidth you ask? Well I guess I don't NEED to very often, but there seems to be an issue with the test servers and how the bandwidth is allocated on the initial test, so I figured asking about it wouldn't hurt. On subsequent back-to-back tests there seems to be a 'ramp up', at least on the download test, that then gives more accurate and expected results. CA3LE 1 Quote Link to comment Share on other sites More sharing options...
CA3LE Posted April 14, 2023 CID Share Posted April 14, 2023 Hi Forsi, welcome to the forum! First, I want to direct you to my beta sign up, simple interact with that topic by voting and you'll be notified when the new beta is released. I would love for you to test my new tools when they're released soon. Under the current version you may get a higher result by using multithread. I see now that you tried that option, maybe also try selecting different mirrors, maybe your route to the NY datacenter has an issue. It's very hard to get 1000 Mbps on my linear test because of how the information loads into the page. You have to really have a lot of things align to achieve those speed using my default test. Here's an example of one of our members @sst3ph3nss recently pulling >900 Mbps using the linear test. Running over 1000 tests in 4 days. Pretty consistent. ?t=u&d=04-08-2023+%2F+04-13-2023&l=1000&q=sst3ph3nss's Speed Test Results It's possible, just very difficult. 38 minutes ago, Forsi said: On subsequent back-to-back tests there seems to be a 'ramp up', at least on the download test, that then gives more accurate and expected results. The new version actually plays on that observation you've made. You'll see. I think you'll like it. I especially want people with connections like yours putting it through its paces. I'd be interested to see your Starlink results as well. Other speed tests don't stream the data in the same way as TestMy.net. Fact, no other speed test does. I'm not sure yet if the new version will be multithread or linear by default. I really like the number multithread gives me but the default linear method I've used pretty much since the beginning seems to be able to passively detect more issues. Time and testing will tell if the new version's multithread is somehow just as special. So give the people what they want to see... a big number -- or give the people what they probably NEED to see -- the result that helps them figure out their issue the best. Side note, if things all line up right, under certain circumstance the linear result can outperform the multithread. A huge disparity between their results can indicate a number of issues. So the new test continues to do both. But in an entirely new way, an entirely new algorithm. Forsi 1 Quote Link to comment Share on other sites More sharing options...
Forsi Posted April 14, 2023 Author CID Share Posted April 14, 2023 I appreciate the reply and all the info! I'm currently using starlink as an automatic failover for the fiber using mwan3, so until I pause the service (because the fiber deployment in my area has been awesome) I'd be happy to give it a go on the beta with starlink as well. WOW though, 1000 tests in 4 days?! haha I was kinda afraid of getting in trouble for pulling too much data with just a few tests, but I guess that's a non issue eh? I'm currently looking for a Pi cm4 board or something openwrt capable with SFP and 2.5gbps ethernet interfaces soon which will give the ability to use the entire 1.25gbps fiber connection. Looking forward to testing that out! Is there anything planned or already in the beta that would show bufferbloat or give the latency changes from idle connection to loaded? There's only one... I guess 'mainstream' option for getting a read on this that I'm aware of outside of some linux utilities that virtually nobody but total nerds like me use xD I am getting real numbers (i.e. total saturation of the connection) to my remote server (screenshots), but 200MB goes so quickly here that I see it spike to 900ish megabit for only a second and the test is done, but showing 500-600mbps. Quote Link to comment Share on other sites More sharing options...
xs1 Posted April 17, 2023 CID Share Posted April 17, 2023 On 4/13/2023 at 10:49 PM, Forsi said: When testing 1gbps up/down fiber connection over wired ethernet via the automatic test on both a laptop and desktop I see bandwidth much lower than I'd typically see when using my own remote server or other speed test utilities. After clicking the 'test again' button 3-4 times the download test seems to cap out at around 600mbps, and the upload test seems capped at around 400mbps. I don't think I've ever seen the expected results on the first run. It seems as though this is skewing test history to show an average of much less bandwidth than I have been able to confirm having through other testing methodshttps://www.waveform.com/tools/bufferbloat?test-id=d6c40a41-56e8-4770-ba94-756ae659094d Why do you need to test a connection that has that much bandwidth you ask? Well I guess I don't NEED to very often, but there seems to be an issue with the test servers and how the bandwidth is allocated on the initial test, so I figured asking about it wouldn't hurt. On subsequent back-to-back tests there seems to be a 'ramp up', at least on the download test, that then gives more accurate and expected results. I'm also awaiting the new beta test, lol. Testmy does well, but the upload sometimes struggles in the current revision. 300 is about what i average here. Granted, i am also in a very over-provisioned area, so my latency & speeds very depending on the rest of the neighborhood.. 🙄 Quote Link to comment Share on other sites More sharing options...
Forsi Posted April 18, 2023 Author CID Share Posted April 18, 2023 On 4/16/2023 at 11:52 PM, xs1 said: I'm also awaiting the new beta test, lol. Testmy does well, but the upload sometimes struggles in the current revision. 300 is about what i average here. Granted, i am also in a very over-provisioned area, so my latency & speeds very depending on the rest of the neighborhood.. 🙄 Over provisioned?! I call that getting what you pay for assuming you're on a 'Gigabit' plan! LOL Sorry, I'm still bitter from my experience with Frontier DSL SQM+CAKE should totally get rid of that +18ms increase when your upload is saturated Quote Link to comment Share on other sites More sharing options...
xs1 Posted April 19, 2023 CID Share Posted April 19, 2023 (edited) On 4/18/2023 at 12:50 AM, Forsi said: Over provisioned?! I call that getting what you pay for assuming you're on a 'Gigabit' plan! LOL Sorry, I'm still bitter from my experience with Frontier DSL SQM+CAKE should totally get rid of that +18ms increase when your upload is saturated lol this test was taken @ midnight. During the day it can get pretty bad around here, sometimes dipping as low as 400/600. >.< Also, other than speed tests, i haven't really found a way to saturate my upload speeds. Frontier (as well as many ISP's) shapes bandwidth depending on the program. P2P/FTP programs ( & ports) will be recognized & throttled to the amount of speed per connection. For example, if a user with a decent connection tries to download a file from me, it will peak at lets say 21Mbs; (hypothetical speed) but if the same user has 6 upload slots to me, they will than have 6x 21MBs. Its very difficult to utilize much of my maximum potential. basically a single vs multithread scenario. Edited April 19, 2023 by xs1 CA3LE 1 Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.