Testing bandwidth throughput without violating the provider's TOS

spammy

I know this is a sensitive topic and most of the providers don't like this idea, so I would appreciate we keep this discussion purely technical.

Basically I have a server that I would like the test the bandwidth throughput on before I put anything important to it. (So my intention is not to "screw" the provider or use it for any illegal purposes), The server has a small HDD size and is NOT in a DMCA-ignored DC which means I can not download all the Kim Kardashian's photo/video and seed it.

I believe a while ago one of the forum members here who was trying to do this by writing an infinite loop to wget some GB-sized file was basically accused of being carrying a DDOS on the network (again, I do not want this to become a debate of who is right who is wrong), so that doesn't seem to be an option as I have the full intention to continue with this service after the test is done.

I have also tried to seed a Debian ISO and guess what? Because there are so many seeders I guess, I had like 30MB of upload traffic in 2 days.

So, is there any way I can test the stability and reliability of the network for a brief period (say even just 48-72 hours) without violating provider's TOS? My home connection is rather slow so I don't think I can just wget a file from the server to my home PC.

Again, appreciate if we keep this discussion technical, if you prefer not to post anything in public, PM should also work.

P.S.: you may assume that I won't go over my bandwith limit even if I run the test at full port speed.

Traffic

Iperf?

ItsChrisG

whos to say that the place(s) you download from OR anywhere across the entire internet path between each other -- is not the one who had the issue?
it sounds like you are too scared of something that you want to exist but doesnt.

a few geographically and network diverse locations testing the speed possible between the 2 locations a couple times over a couple days and a few different times -- would be more than sufficient.

daily

I would get your provider's consent first, if you don't already.

tr1cky

Just go with linode, ~1.2gbps all day.

Edited by @jarland: If you want to make a thread about William, make a thread about him. I will honor the OP's wishes and keep this thread clean of it.

jar

@dailymc said:
I would get your provider's consent first, if you don't already.

This. Legitimate usage is one thing but maxing port speed for 72 hours just to test the network is going to come across a lot better if you're up front with them about it first.

rincewind

Try short bursts (say, 5 minutes) of traffic. Repeat every hour at random offsets - say 6:05-6:10, 7:34-7:39, 8:12-8:17 ....

It takes a few seconds for the TCP connection to figure out good window sizes and hit congestion control to reach peak speeds, I'm guessing 10 to 30 seconds if you have bad sysctl settings. So start measuring only after the first minute or so. That way you only measure steady state behavior.

As long as you space out your bursts, and keep it between a single transmitter/receiver it shouldn't impact your host much (I guess). You can experiment with different pairs of hosts for each burst, and then compute whatever stats you want - geographical, time-of-day etc.

daily

@jarland said:
This. Legitimate usage is one thing but maxing port speed for 72 hours just to test the network is going to come across a lot better if you're up front with them about it first.

Exactly. I don't mean to incriminate you yet, but if you're really interested in keeping this provider after this test, then you want to make sure they are still interested in keeping you as a client. If you can't be reasonable about it, then I have my doubts. But right now it is a fair playing field and I have nothing against you as long as you're willing to be responsible about it.

spammy

@rincewind said:
Try short bursts (say, 5 minutes) of traffic. Repeat every hour at random offsets - say 6:05-6:10, 7:34-7:39, 8:12-8:17 ....

It takes a few seconds for the TCP connection to figure out good window sizes and hit congestion control to reach peak speeds, I'm guessing 10 to 30 seconds if you have bad sysctl settings. So start measuring only after the first minute or so. That way you only measure steady state behavior.

As long as you space out your bursts, and keep it between a single transmitter/receiver it shouldn't impact your host much (I guess). You can experiment with different pairs of hosts for each burst, and then compute whatever stats you want - geographical, time-of-day etc

Like this idea a lot!

AnthonySmith

Well I assume your talking about a dedicated server here otherwise this is kind of a pointless discussion, on that basis ask the host for a test file on the same physical segment, and grab that from multiple sources every 15 minutes running an MTR before hand.

Ikoula

Hello,

i agree with @AnthonySmith you should ask your provider, first he will not say he was not aware and will not accuse for anything and then maybe he can provide you a looking glass for example on the same physical segment.