JON HONEYBALL
Jon solves the mystery of slow Wi-Fi on his Windows laptop, gets annoyed by label-printing software and then kicks back with Flight Simulator
Jon solves the mystery of slow Wi-Fi on his Windows laptop, gets annoyed by label-printing software and then kicks back with Flight Simulator.
iPerf3 is one of those industrystandard tools for measuring network performance. You run it at each end of a connection and then fire it up to determine how quickly the data gets sent over the wire (or Wi-Fi). It’s been around a while and is very much a known quantity.
The app is usually controlled from a command line, which gives a good indication of how hair-shirt it really is, but the core technology is available on just about every computing platform out there. Some developers have wrapped a graphical user interface around the core code to make it more friendly, but being a keyboard warrior, the command line with its endless switches and options is The One True Way.
Well, most of the time. I use Magic iPerf on Android, which works well despite its somewhat leggy status. On macOS, I’m liking iPerfUtil with its native M1 support and it now supports the new Streams (-P) option too.
A Windows version continues to be an issue. If you Google iPerf, you’re pointed to iperf.fr, which appears to be the quasi-official site for all things iPerf. It has links for just about every platform out there, but on Windows development stalled at version 3.1.3 back in 2016. Not surprising: iPerf 3 does what it does and there hasn’t been much need to change it.
If you go to the downloads section and pull down the Windows 64-bit version, you will get a ZIP file. Inside that is the iPerf 3 executable along with a DLL file called cygwin1.dll.
Cygwin is an open-source interface layer, specifically designed to act as a thin translation layer between Linux-style network calls and the Windows API stack.
I was getting inconsistent results with the Windows version of iPerf 3. When connected over 10Gb Ethernet I could get a solid 4-5Gbits/sec throughput, so although it wasn’t hitting the bump stops of the expected 9.4Gbits/sec, it certainly wasn’t shabby. Maybe this was the Intel 10GbE drivers for Windows choking on the traffic levels? Connecting a Windows laptop with the latest 802.11ax chipset through to a hot Wi-Fi 6 router would give good performance on various devices, but Windows laptops with Wi-Fi 6 seemed throttled. I put it down to the various horrors of buggy firmware that I was seeing. My Dell XPS 15 laptop with the Killer AX chipset seemed to get new drivers and firmware every five minutes.
However, I started looking at other tools. Although iPerf 3 is still well respected, and built into other toolsets that I use such as NetSpot, it’s old. Worse still, tools like Speedtest, made famous by Ookla amongst others, take a different approach – there’s no client app, just a modern web browser that connects to a server web process. This seems to be a common currency, sometimes wrapped up in the shell of an app, to measure throughput and maybe this was the way the world was moving.
I downloaded the OpenSpeedTest toolkit ( openspeedtest.com) and installed it onto my Dell desktop test server box. In the meantime, I had also installed Linux Mint onto the Dell, taking the opportunity to repartition the live Windows partition to make space for Mint, thus allowing me to dual boot. Having gone to Mint, I found that the Dell test s e r v e r c o u l d d e l i v e r u p t o 9 .3G b i t s / s e c on iPerf 3 from a macOS client running on a MacBook Pro. This was connected to the 10GbE network backbone via a Thunderbolt dock that comes with a 10GbE adapter. Clearly something was amiss with the Windows stack. This was confirmed by using OpenSpeedTest tools, which would cheerfully give me full speed.
Much debugging ensued. I was doing over the air packet capture using AirPcap to see if the 802.11ax signal was going awry on Windows. Nothing seemed to make sense.
Then I discovered a deep and dark posting from a Windows core engineer at Redmond that revealed the issue was actually inside Cygwin, and that this was throttling the throughput. There was a new API for automatically assigning the sockets on a net call like the one iPerf was making, and that the version of Cygwin held within the ZIP file hosted on iperf.fr was well out of date. All I had to do was to go to the Cygwin site and download the latest version of the file, use that and all would be good. Which turned out to be the case.
So I had been banging my head against the wall with Windows iPerf 3 because the ZIP file on iperf.fr used a Cygwin file that is out of date. Both components are open source, but
unless someone joins up the dots, it can be difficult to work out what is going on.
Now I have 802.11ax Windows laptops delivering proper iPerf 3 speeds. And I’m really liking the OpenSpeedTest platform too. At least on the Linux version, it comes as one self-contained package, and you just point your client web browser at it on port 3000 by default. iPerf 3 is still useful, especially if you want to run data testing over many minutes or even hours. OpenSpeedTest is more of a quick one-shot tool. Nevertheless, both have their place, and iPerf3 is useful on Windows once you get the right DLL in place.
Whole man-weeks have disappeared on this journey, but that’s par for the course, it seems. It’s been a good reminder to check and double-check the veracity of the sources. In my journey, I discovered that iPerf 3 has been split off with a splinter group pushing forward with versions after 3.1.3; head to software.
es.net/iperf for this group. It doesn’t provide binaries, assuming that you’ll compile your own. But in a somewhat cruel twist of fate, its site has a page covering binary distribution that states “Windows: iPerf3 binaries for Windows are available from iperf.fr
(and other sources)”. So it’s pointing me in the direction of iperf.fr! And going back to iperf.fr would take me full circle to the version that has issues. I think I need gin, and lots of it.
Label printing software
This month has seen a raft of old and dying laptops crossing my path. It started with Mike’s tired old Toshiba laptop, which I’ve resurrected several times over the past decade. This time it decided that it would take over five minutes to boot into Windows 10. My initial suspicion that it was malware at work proved to be false. The computer was clean, and pretty much up to date. There was nothing obvious in Task Manager, and a few sweeps of the computer showed nothing untoward. Which might point the finger at a failing hard disk, which is old enough (at over a decade) to have a spinning rust platter for storage.
I then did something rather silly: I ran Windows Update. This ran with no issues, but after a reboot all I got was a blank screen with no login box.
I thought the display driver might be the issue, so I tried logging in blind, by just typing in the password. That didn’t work. So I had to resort to firing up Safe Mode. It has been a while since I have had to recover a Windows 10 machine, as I keep most of my Windows 10 installs inside virtual machines, for all the benefits that brings.
So I had to remind myself that the way to get into Safe Mode is to boot three times in succession, interrupting it in the process. That worked, so I took a spare external SSD and copied off all of the important data; family photos and files that obviously have much value to Mike. I think he has a backup on an external USB drive, but it was best to be sure.
After a few hours of fiddling, I gave up. I returned the tired old Toshiba to Mike and told him that he really ought to buy something new. He went to Dell and ordered up a low-end spec laptop cost about £350. Fortunately, he took my recommendation of going with 8GB of RAM rather than the default 4GB, so hopefully this will have a long life ahead of him. It might not last a decade, but at this price it’s not expensive computing viewed over a number of years.
And then Gemma’s old MacBook Air died. Gemma works for Manfood (@welovemanfood), our pickles and condiments company. She does paperwork on it – production
“The laptop’s hard drive is old enough to have a spinning rust platter for storage”