requesting help with network speeds on G4 Quicksilver OSX 10.4.11

[bigmessowires]

My new G4 Quicksilver running 10.4.11 seems to have strangely slow network speeds, but I don't understand why. The Quicksilver has a gigabit Ethernet port, where I've plugged in the very same Ethernet cable that's normally attached to my desktop computer. The desktop reports speeds around 450 Mbps in speed tests, but the Quicksilver with the same network connection only manages about 2 Mbps (250-300 KB/sec) downloading large files from the web or FTP sites. I've tried a few different sites, and tried downloads via InterWebPPC and also via Fetch. The Network tab of Activity Monitor shows a nice consistent download speed of 250-300 KB/sec.

I've also tried transferring files over my home network between the Quicksilver and my MacBook Air when both are connected to wired Ethernet. The results are faster on average, but extremely variable. Activity Monitor on the Quicksilver shows data received per second from as high as 25 MB/sec to as low as 30 KB/sec. It'll go fast for a few seconds, then slow hard, then pick up again later, while nothing much else is happening on the home network. Transferring a multi-gigabyte file, the average transfer rate was about 3 MB/sec or 25 Mbps, which is better than internet downloads but still much slower than it should be.

I've not experienced any network speed problems with my modern computers, and I typically get symmetric internet speeds of 400-500 Mbps.

The network setup is relatively complicated, which may be where my problem lies. The gigabit fiber entry point is connected to a Ubiquiti EdgeMAX EdgeRouter X gigabit ethernet router. Some devices are wired directly to that, while some are connected to a hub or switch that's plugged into the EdgeRouter X. There's also an eero mesh base configured in bridge mode, but that only serves WiFi devices and I don't think it's relevant here. There's also an Airport Extreme with gigabit ethernet connected to the wired network, configured in bridge mode, which is normally only used for a Time Machine network drive. I did some tests while the Quicksilver was connected to my wired desktop cable, and some with it connected to the Airport Extreme, with similar results.

Any guess where to start looking? This feels more like a protocol problem than a low-level connection speed problem, but I don't really know.

EDIT: From the Quicksilver I just now tried the speed test at speed.googlefiber.net, and it reported 123 Mbps down and 225 Mbps up, much higher than the actual speeds I seem to be getting from the web.

 

[Phipli]

Activity Monitor on the Quicksilver shows data received per second from as high as 25 MB/sec

How fast is your hard disk again?

 

[Phipli]

From the Quicksilver I just now tried the speed test at speed.googlefiber.net, and it reported 123 Mbps down and 225 Mbps up, much higher than the actual speeds I seem to be getting from the web.

What is CPU activity like during various activities? During downloads, during speed tests, FTP and appleshare?

Depending on protocol, you'll be capped by the hard disk, CPU or the server.

 

[Forrest]

Remember your G4 only has a single CPU core - doing more than 1 thing at a time will slow it down. Wasn’t a problem back in the day, unless you were benchmarking. I think we’re all just spoiled when even a $50 Raspberry Pi computer has 4 64 bit CPU cores at 1.5 GHz or faster.

 

[davidg5678]

Are you trying to run a modern speed test like speedtest.net? There is a ton of overhead in actually running the javascript and everything else online, so it wouldn't surprise me if the problem is that the G4 is struggling to cope with overhead, rather than actually having slow ethernet. The modern internet is just a lot for a G4 processor to handle, and I think that is really what your speed tests are showing more than the ethernet throughput capabilities.

File transfer speed tests will show the network abilities of your computer, but also the hard disk bandwidth limitations, so you are really testing two things at once, which makes it less accurate as a test of ethernet.

The solution I worked out to test the theoretical max internet speed on my iBook G3 Clamshell was a command line program called iperf2. (I don't believe iperf3 runs on PPC, but iperf2 works fine.) This program doesn't rely on running the modern internet, nor is it capped by your hard disk speed. iperf2 needs to be downloaded to both your G4 and another modern computer in your house, and then the two computers will send data back and forth.

From what I recall, my iBook scored somewhere in the 70mbps ballpark, with a 100mbps ethernet jack. I'm sure your G4 will blow it out of the water... https://iperf.fr/iperf-doc.php

 

[bigmessowires]

How fast is your hard disk again?

File transfer speed tests will show the network abilities of your computer, but also the hard disk bandwidth limitations, so you are really testing two things at once, which makes it less accurate as a test of ethernet.

Right, good thoughts. At the moment I've gone back to using the IDE-to-SATA adapter, which has a max speed of about 55 MB/sec when writing large blocks (roughly 1MB), and maybe 10-40 MB/sec when writing smaller blocks. So you're right, there's no way I'm going to see transfer speeds of hundreds of megabytes per second. This would explain why local file sharing speeds never went above 25 MB/sec even briefly. But I'm still not sure why the average speed was about 3 MB/sec and sometimes the speed dipped to under 1 MB/sec for a while. Maybe there's some transfer speed negotiation happening at a low level?

Come to think of it, all my local file sharing tests involved copying a file to the Quicksilver, never copying a file from it. I'll try that.

But disk speed can't explain why internet downloads are only 300 KB/sec. That must be something else.

Remember your G4 only has a single CPU core - doing more than 1 thing at a time will slow it down.

Right, I'd thought about this. I wasn't running any other applications on the Quicksilver except Activity Monitor. Perhaps there's some required processing of each downloaded block, and the time needed to do that swamps the actual network transmission time for the block. That's why I tried Fetch and local file sharing as well as InterWebPPC, hoping that one of them might involve less overhead. Maybe there's a simpler protocol I could try?

But in the end, I'm dubious that a 933 MHz single-core G4 can't receive and store data from the network at more than 300 KB/sec (2.4 Mbps). If that were the case, it would have seemed pointless for Apple to put a 1000 Mbps Ethernet port on the computer, or even 100 Mbps.

Are you trying to run a modern speed test like speedtest.net? There is a ton of overhead in actually running the javascript and everything else online, so it wouldn't surprise me if the problem is that the G4 is struggling to cope with overhead, rather than actually having slow ethernet. The modern internet is just a lot for a G4 processor to handle, and I think that is really what your speed tests are showing more than the ethernet throughput capabilities.

I agree, this is definitely a factor. I was using speed.googlefiber.net because it was one of the most lightweight speed testers I could find, but I'm sure the Javascript overhead is still significant. I was interpreting the reported numbers as a lower bound on the true speed. I'll take a look at iPerf2.

At the end of the day, I'm really just looking for a way to download software from the web or FTP sites onto this G4 at faster than 300 KB/sec. Even with the hard disk bottlenecks and relatively slow G4, shouldn't something in the tens of MB/sec be possible? Could this be an issue with Ethernet settings, MTU size, IPv6, half/full duplex, something?

 

[Phipli]

At the end of the day, I'm really just looking for a way to download software from the web or FTP sites onto this G4 at faster than 300 KB/sec.

There is something not right. My beige Macs download over FTP at faster than 1 MB/s over LAN.

 

[bigmessowires]

Over LAN - from a server on your home network? Maybe that's a clue. Have you tried downloading from Macintosh Garden via FTP? Or other sources on the web?

 

[herd]

I don't have internet that fast, but local file transfer between my machines seems to be the right speed. Activity Monitor can show what the CPU is doing in addition to the transfer speed. Whenever I notice low benchmark results, Activity Monitor usually reveals something like spotlight indexing as the problem. You can also use firewire as a network connection, and it seems to use less CPU than ethernet.

What happens when you turn off IPV6?

 

[Phipli]

Over LAN - from a server on your home network? Maybe that's a clue. Have you tried downloading from Macintosh Garden via FTP? Or other sources on the web?

Website speeds are arbitary, even on a modern machine. Macintosh Garden doesn't saturate my laptop's download. I should be able to get about 8MB/s but rarely do. Steam is one of the few things that saturates my connection.

I see that kind of speed when using 10baseT, downloading from my phone's FTP server. It usually reduces with time, but I think settles at about 800kB/s on my 7200. It goes faster if I put a 100baseT card in it.

Machine isn't set up ATM - I brought it to the meetup and today has been... Well. I haven't done much. Just resting. I'll see if I have an old photo of it transfering from last week.

 

[Phipli]

Over LAN - from a server on your home network? Maybe that's a clue. Have you tried downloading from Macintosh Garden via FTP? Or other sources on the web?

This is my 400MHz G3 Pismo, using an 802.11g WiFi card in the CardBus slot, from the CyberDuck FTP client in 10.4 :



Since I took the photo, it has increased to 1.4MB/s stable which is about 11Mbit/s. This aligns with 802.11b speeds, I'm not sure how to tell if it's working in that mode.

Edit : Connection was 802.11g (54Mbit).

 

[Phipli]

I plugged in the (100baseT) ethernet. I got a peak of 3.2MB/s, in CyberDuck, with the interface traffic showing as about 4.3MB/s peak, so about 35Mbit/s.

CPU usage was high, but not saturated.

 

[Phipli]

And testing downloading from Mac Garden (Same 400MHz G3 Pismo, 100baseT) peaks at 1.2MB/s and completely saturates the CPU.

So in my instance, even with only 100baseT, the CPU is likely the bottleneck.

 

[LaPorta]

I can try to run comparison tests on my QS if you tell me exactly what you treated and how. That being said, it likely wouldn’t be a fair comparison since mine is a DP model.

 

[Phipli]

Another Last quick test - I rebooted in 9.2.2.

Using the FTP client Transmit to download a file from an FTP server on my phone with the Pismo using ethernet, it peaked at just above 2MB/s (photo shows just under 2MB/s).



Edit : other FTP transfers have broken 3MB/s. Not sure why the first was slower.

 

[Phipli]

And Classilla :



Speeds changed a bit for each download and generally start higher and slowly reduce.

 

[bigmessowires]

I think this is an issue with Macintosh Garden's server configuration interacting weirdly with the Quicksilver's Ethernet settings or something in OSX 10.4. I'm normally in the habit of clicking the first download link in Garden listings, which is a www server in Sweden. On my MacBook Air this provides speedy downloads, but on the QS the same link gives me 300 KB/sec. The second link is an FTP server in Sweden, which also gives 300 KB/sec. The third link is a US mirror, which strangely does not work at all from either the MBA or Quicksilver. But the fourth link is www mirror in Germany, and that gave me download speeds of 8 MB/sec on the Quicksilver!

I tried this on a few different Garden titles with the same results. The German mirror provides dramatically faster download speeds than the default Swedish link. But on a MBA, both links are very fast.

I then tried another web site with large dummy files you can download for testing purposes, and a public FTP server, and both gave me speeds about 2 MB/sec.

Maybe there's a peculiar interaction between the Garden's Swedish server and InterWebPPC, Fetch, or my Ethernet settings under 10.4.11. Sorry for the (mostly) false alarm, I thought I'd tested other web sites to confirm this wasn't a problem specific to Macintosh Garden, but I didn't test thoroughly enough.

 

[cheesestraws]

Has the MTU on the ethernet interface got changed on the G4 somehow? Have a look in network preferences or ifconfig...

 

[Phipli]

I think this is an issue with Macintosh Garden's server configuration interacting weirdly with the Quicksilver's Ethernet settings or something in OSX 10.4.

This is unlikely - I was testing in the same browser, same OS and a slower computer. These are already vintage machines and not exactly fine web browsing machines.

Only thing I'll say is make sure you're using the G4 specific version of the browser, it has optimisations. I'm using the G3 specific version for my G3.