Huge variation in download speed

[boomerang]

Here's the background. I installed a two router solution for a friend about, maybe six to eight years ago. (Neither of us remembers exactly how long ago.) One in the house, one in the shop about 300 feet away. Two, what were at that time high-end Netgear residential wireless routers. It has worked flawlessly all this time with a wireless connection between the two. Lately, they've been getting slow and the transformers are failing right and left.

He decided to up his speed tier so Comcast came out and installed one of their gateway devices as his Modem was EOL. They also wanted Comcast's triple play with phone. In the house, the wired computers are getting 80Mbps/12Mbps which is slightly faster than the tier he is paying for. Wireless is great too. The Netgear's are out of the mix now.

Out to the shop, he had run a cable about a year prior. Last week I helped him get that hooked up at either end. He's a hard-headed guy and wouldn't listen to me. He didn't want to pay for the right cables he bought a single wrong cable. Four pair phone wire. The gauge is too small, the twist is most likely not at spec for network usage and to top it off, he was already using one of the pairs for the phone out there. So we put ends on four of the wires, enough wires for a 100Mbps connection (based on some research) and had the continuity we needed. Multiple speedtests pretty consistently show 9Mbps/12Mbps out at the shop. The pair he's using for the phone was disconnected during the tests.

I've told him to take a laptop out there, swap the wire over to it and see what he gets to see if the problem is his desktop in the shop.

If he gets the same results with the laptop, notwithstanding all the really wrong aspects of this wired connection, why such good upload and such bad download? I would have suspected more linear results. Slow on both and somewhat proportionately. The download is actually fine for his needs. He's paying for a tier way, way faster than he needs. But we're both kind of curious why the results are so heavily skewed towards the upload.

Also thinking about a Ubiquiti solution which will have to wait until the spring as I'm about to vacate the frozen north for the winter and it will have to wait until I get back. I've got to do some more research on these to figure out the best solution for his situation. Maybe these? http://www.amazon.com/Ubiquiti-Nano...8&qid=1450612716&sr=8-1&keywords=nano+loco+m2

Sorry this is so long.

 

[Gryz]

If your cables are not up to spec for the technology you are using (100 Mbps Ethernet, or 1 Gbps), then all bets are off. The only real solution will probably be to put a proper Cat-5E or Cat-6 cable in the ground. Until you do that, you might mess around for a year, and never get it working to your satisfaction.

I guess your main concern is not speed, but packet-loss. The bits over the wire will run either 100 Mbps all the time, or 1 Gbps all the time. Ethernet is not like wireless or DSL, the speeds don't adjust on the fly. The fact you are seeing different "goodput" speeds, is because packet-loss has a huge impact on goodput. Rule of thumb used to be: 1% packetloss with TCP causes a 50% performance degradation. (Goodput is the actual throughput that is seen by the applications/users. You might run your network at 1 Gbps, but if the application only sees 1 MegaByte of data being transferred, then the "goodput" is only 8 Mbps).

Try to see if you can see the statistics for packet-loss. Either on the endpoints you are testing (the PCs or servers). Or on the routers/switches/basestations in between. On Unix this is easy. I have no idea how to look at packet-loss numbers on Windows or Macs.

Make sure you know what you are testing. Most people connect to a "speedtest" webpage, and believe the numbers the speedtest is telling. These numbers give an indication of something. But for real troubleshooting, they are useless. If you want to test your friend's cable, set up a test that tests only that. And not your internet connection.

Many people make the mistake that when they test their local networks, in fact they test the speed of their harddrives (or their SSDs, or their busses). Use software that doesn't just copy a file from disk to remote disk. Most disks are slower than 1 Gbps. Some disks (especially external disks, like USB-disks) might be even slower. Your test application should copy from RAM to RAM. Also, doing file copies over a network protocol (like Samba/CIFS/SMB) will be (a lot) slower than just sending bytes from RAM to RAM via TCP.

I'm sure others here will know what the best free network-speed test application is.