HOMELAB: A definitive guide to using iPerf3 to measure network performance! (Part 1)
ฝัง
- เผยแพร่เมื่อ 10 ก.พ. 2025
- In this first of two videos we will explore the use of iPerf3 in measuring network performance using both TCP and UDP.
Part 2 of this series is posted here: • HOMELAB: A definitive ...
The GameSpy mug took me back to early 2000's and C&C Zero Hour! Wow. Cool vid!
The only guy that can wear two watches and not look stupid.
I came here to learn something about iPerf3 and got a wealth of information about networking I never thought I'd understand for a while. Will be rewatching this a number of times. Look forward to the next video!
Gamespy mug!!! Throwback.
Oh yea.. back in the day I ran one of the better west coast Quakeworld servers. Fun stuff!
Beautiful details given on dropped datagrams and why because of buffer_full. I learned a lot again . Love it.
Great video Jeff. I have been using Iperf3 for a while and completely missed the --bidir flag. You just cut my test time in half 🙂
Extremely informative video.
I needed this video! Thank you so much!!
Thanks for sharing good knowledge Jeff.
@16:06 That's what she said.
But seriously..This was a great explanation of iPerf. I've used it in the past but never did much with the different flags.
Great video, Jeff! I didn't know about the parallel flag. I'm gonna have to try that out.
iperf helped me also test multicast stuff, really cool tool.
I sometimes wish I could just run iperf on routers and switches directly, it would make testing performance and finding bottlenecks a bit easier! Very useful tool
Yea, there are iPerf binaries for MIPS based devices, so you can run it on some device, but it would be great if it was available on things like Cisco switches. It is available on Ubiquiti switches.
You can run it on Cisco IOSXE routers and some IOSXE switches (not lite versions). It uses a Cisco feature called Guestshell which loads a Linux VM/Container. It takes a bit of configuration but does work.
@@dsmidt Ah excellent I'll give that a try.
Newer Fortigate versions support iperf as well. It's called something different but uses iperf under the hood and can be used against a standard iperf3 client on the other side.
The text in the thumbnail is cut off on mobile devices, maybe you can position it a little higher.
Iperf3 is really useful. I'm currently using it to measure network impairment effects in various cloud setups. Between different locations, providers, VMs with guaranteed and shared resources, internal and external routes, bare metal vs virtualized, local to cloud etc.
Thanks for the overview!
Ah yea, thanks for catching that!
One thing I wish I knew about sooner was the ability to run iperf3 in a CLI on an Android smartphone. It's so much easier than carrying a laptop around a building or outside, even works with USB Ethernet adapters and of course, you always have your smartphone with you. Even done 2.5 Gb/s with it, modern smartphones are crazy.
I'll also mention when I need to use iperf3 on Windows I use it in WSL, works a treat, although I only have 10 GbE stuff to test.
Those UDP limitations are very interestingly, never had the need to test it but now I want to just to see how it will cope.
Oh yea, running it on a phone would be a great tool for debugging!
Can I ask what the title of the blogpost was that you were talking about at 4:36?
Was it "Three Reasons Why You Should Not Use iPerf3 on Windows"?
Yep, and if you read the update they mention the version issue is partly resolved, but the Cygwin layer of course can't be. They suggest nttcp, which does work well on Windows, but sucks on Linux.
Did you see that job opening in Houston? Vice President of Technology (Space Centre)
I think MS against iperf3 is because it exposes a ton of Windows issues when running 40+ gb/s. Lol. However, I did manage to get about 38gb both directions across a 40gb link to my nas. It took a ton of tweaking and musical pcie slots.
The musical PCIE slots is a real thing - and especially because it is hard in windows to know how something shows up... lspci in linux is very very helpful!
Hey Jeff, why were the total datagrams count on the sender and receiver side different, but no lost datagrams counted?
Excellent eye - I will answer that in part 2!
@jeffsponaugle6339 Great look forward to it!
Just posted part 2: th-cam.com/video/TRE8yxKoLG4/w-d-xo.html
did I miss something? I can strongly remember a video where Jeff explained how he designed the house with all of the infrastructure in mind and I can not find it. Am I blind or did he take it down?
btw., we also sometimes use iperf on the mainframe at work, nifty little tool!
@@_-Karl-_ Ohh nice, thank you! I thought there was a deicated video that somehow went missing but are absolutely right, it was part of one of the videos you mentioned
So does that mean jumbo frames only affects UDP and not TCP?
No, but the effect in iPerf3 is much greater with UDP due to the individual system calls per datagram. TCP transmission rates will be slightly higher with jumbo frames due to the overall efficiency.
It could be fun to do a video on a beginner homelab setup with ebay parts: a basic NAS, plex server, networking, and maybe a vm. Hoping to setup some of that this year myself, but I'm pretty new to all of it.
Casually wearing TWO smartwatches 😅 Btw you still looking for an editor with IT background? I know a guy.
Yea, it is annoying that Apple took away the pulseox stuff because of patent problems!