All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
RS customers - I need some help!
Mods - I'm not sure where this belongs, move it as necessary.
So, we've been working hard on making some network changes....because our existing network setup has been showing weird speeds with speed tests. We've tested this internally and with a couple customers, but I need a larger test group to see if any further tweaks are necessary.
Please read this before posting a speed test:
I would prefer that you use a physically close server to where you server is located (don't rely on automatic choosing, it's not always accurate).
If your speed test is less than exciting, I may reach out to you via PM for further diagnosis (or I may not, who knows). Make sure to take into consideration your actual current usage will limit the results you see in speed tests.
Please please please keep this on topic. This is not a support hotline, I will not answer anything unrelated to my request here. You may private message me with any questions you may have unrelated to this.
Speed test results from RS customers ONLY please.
I'm only concentrating on upload speeds for these specific tests. We don't limit inbound traffic, so you'll most likely see better download speeds if you have a 10G port. Please don't ask me how you can get a 10G port, you either get lucky with a motherboard that has one onboard or if you committed to an unmetered option that requires a 10G port upgrade (downgrading will remove the 10G port).
To save everyone some time, here are a few commands for Linux:
iPerf3 (via YABS):
curl -sL yabs.sh | bash -s -- -g -f
Speedtest.net (Ubuntu/Debian):
sudo apt-get install curl
curl -s https://packagecloud.io/install/repositories/ookla/speedtest-cli/script.deb.sh | sudo bash
sudo apt-get install speedtest
Speedtest.net (CentOS/Redhat/Alma/Rocky)
curl -s https://packagecloud.io/install/repositories/ookla/speedtest-cli/script.rpm.sh | sudo bash
sudo yum install speedtest
Here are some speed test server suggestions, you are not obligated to use them:
Miami: speedtest -s 12215
Los Angeles: speedtest -s 19230
New York City Metro: speedtest -s 18956
Comments
Look at the love guys. I meant speed.
MIA https://www.speedtest.net/result/c/ac3f1165-0815-4fc4-866f-adc13441374e.png
LAX https://www.speedtest.net/result/c/44b20739-420b-4ba5-ade2-335253f52835.png
Haha thanks for being the original guinea pig on this one.
I have like two servers with you I can test later. Will get back.
Speedtest by Ookla
nice
I have one at NYC metro.
Not going to lie, that 5 Gbps download from half way around the world is no joke. I had to do a double take there.
Miami (with bbr):
Awesome, I'll see if we can re route to uztelecom
NYC Metro
LAX
So I altered the test slightly by using your LAX test file (from the Miami server above):
not as good as a server in PHX (which is expected), but the PHX server gets 320 MB/s for above test file from your LAX origin while this - yours - Miami server has 50MB/s which is ok-ish.
But: I also have a NYC server with you, and that is way worse:
So while this is a 1G nic in NYC (and not 10G as in Mia and Phx) I would still love if you first look into upload speeds from the US (e.g. West Coast) to e.g. NYC or inter-DC speeds.
My NYC server is looking much better
@bgerard yes that looks much better and would be usable with the numbers you get.
Sry, did I mixed it up again? Whenever someone (or even those tests) talk about upload/download or send vs recv speed, I mix it up. Can't we talk about egress and ingress?
E.g. yabs: send=egress and recv=ingress.
So when you said "upload speeds" you meant egress from the server?
Edit: just checked, my NYC server of yours did not had bbr configured, this looks much better:
NYC with bbr:
So it would be good to ask if the testers use bbr or other congestion control algos. Or did you changed your routes from West Coast to East Coast?
Thanks for the super detailed analysis here. This specific scenario isn't what I'm looking to test here as I only really care from the server to as close to our network edge as possible right now. The changes we made are strictly with localized switches within our network, which is why I asked for closer locations to be tested.
Route optimization is something we'll re-introduce at a later time. We have a pretty complete product, but after running it in LA for a couple of months, there are some optimizations that we thought are absolutely necessary before we go live on a larger scale.
So I cannot really do some exhaustive speedtests now as your IP adresses are not properly geolocated. Ookla seems to be using Maxmind to geolocate the requester for it's "--list" request and by that - as you say yourself - the "nearest" speedtest servers are not really near.
Now it's pretty easy to correct these geolocations of IP addresses at Maxmind, unfortunately it takes about one to two weeks to see results (depending on how fast ookla is updating the data from e.g. Maxmind).
I'd still highly recommend to correct the locations of your ranges (in LAX or NYC or MIA) into Maxmind as all your customers that use standard tests (that rely on speedtest by ookla) would get better "near" results.
E.g. with my script where i can test dozens of nearest locations via speedtest, I'm right now testing the state of Kansas for your LA location (as the ip address is only geolocated as "US" and that is centered in the US in the middle of nowhere in Kansas).
That won't work
And speedtests fallback for direct tests (using ping to deem a server nearby) does not work as well, that situates a LA server of yours somewhere between Seattle and Canada...
You can specify a server with the -s command. Here you go: https://williamyaps.github.io/wlmjavascript/servercli.html
But again, if you were able to push your purchased bandwidth outbound, then you're good to go, all I care about right now (from your server to the edge of our network). Beyond that is a future project.
Will check later
Hey Radic,
The servers and deals are superb . I have impulsively bought 2 servers lol . While the NYC metro gives me 1 Gigabit up and down out of the box , the LA server is giving me < ~ 100 Mbits in uplink . I turned BBR and that fixed it . I don't know if it's something i should always consider turning on as i have never done so for any providers .
The server is rock solid though , i am just worried if any hiccups/low speed is encountered by my end users .
Thank you
@sk7411 from where did you test these? I get ~1Gbit up in LA with a nearby location. 100Mbit sounds like testing from the east coast (with a LA server) as you said NYC gets you 1 Gbit.
E.g. here is a speedtest (LA server, bbr turned on) (city|up|down|ping):
As you see I got 935Mbit/s up and 940Mbit/s down in Los Angeles.
@OhJohn , it was from my LA server . I coul only see Dallas and LA during the yabsh test. And it gave me around ~800-900 Mbit/s for all of the regions including some European region. This is ofcourse after i turned the BBR on after i read your comment .
I was always under the impression that network should just work right out of the box without tuning it for bandwidth. I want to understand if turning BBR has any impact on the latency or have any downsides . The server is running smooth though! Love it!
@MrRadic
Notes: Horrible upload to Windstream in all US locations
MIA speedtest:
Route optimization outside of our network is a future project.
@nullroute ,
Enabling BBR fixes it .
root@debian:~# speedtest -o la02.speedtest.windstream.net
Speedtest by Ookla
Idle Latency: 76.36 ms (jitter: 0.02ms, low: 76.35ms, high: 76.40ms)
Download: 925.94 Mbps (data used: 1.6 GB)
124.09 ms (jitter: 39.34ms, low: 79.84ms, high: 173.12ms)
Upload: 917.16 Mbps (data used: 1.3 GB)
114.40 ms (jitter: 34.50ms, low: 76.55ms, high: 152.82ms)
Packet Loss: 0.0%
Result URL: https://www.speedtest.net/result/c/b10631d0-bcbb-4adf-b37d-d46f63917aad
root@debian:~# speedtest -o atlanta02.speedtest.windstream.net
Speedtest by Ookla
Idle Latency: 21.25 ms (jitter: 0.04ms, low: 21.23ms, high: 21.29ms)
Download: 941.27 Mbps (data used: 1.1 GB)
30.45 ms (jitter: 0.67ms, low: 22.16ms, high: 32.53ms)
Upload: 939.36 Mbps (data used: 1.1 GB)
23.13 ms (jitter: 0.56ms, low: 22.19ms, high: 25.53ms)
Packet Loss: 0.0%
Result URL: https://www.speedtest.net/result/c/5661efbd-1e11-4c76-8fa2-2854ac4bfa2c
Can you please provide me with the best way to do this?
Sure, If you have OS kernel > 4.9
you can nano /etc/sysctl.conf and put
net.core.default_qdisc=fq
net.ipv4.tcp_congestion_control=bbr
After that , run sysctl -p
You rock, thanks.
NYC Metro
Speedtest