Fiber connection behaving odd in wireless vs cable - seeking input from someone with experience

V

Vaske

n00b
Joined
Mar 10, 2016
Messages
23
I have a fiber connection and I notice that I am not getting the speeds I pay for. I have tried three different computers running windows 7, windows 10 and Ubuntu (linux). I have switched all the networking cables and upgraded them from cat 6 to cat 7. I have switched out the router and tried three different ones. I have eliminated all points of failure on my side of the network and yet, the speeds make no sense.

I am not getting the upload speed I pay for which is 250 mbit. I pay for 250 / 250 and I get 230-250 down and about 92 mbit up.

When testing with a laptop or a mobile phone using the WIFI of the routers, all three routers, different brands, the upload speed is higher. A lot higher.

Does anyone have any clues why wifi would give higher upload speeds than using a cable ?

I've had the ISP here three different times and they witnessed it as we swapped out everything. Then they take in their own router and measure higher speeds with a different router using the wireless holding a phone 2 inches away from it.

However everytime we measure with a cat 6 / cat 7 cable the speeds caps out at ~90 mbit up.


I have a hypothesis:
A laptop / android phone using WIFI sends the signals to the router. The router then transmits the data to the fiber modem and out on the net. Essentially the router works as a booster of kind.

Whereas when the connection goes through cable, the data gets routed through the router because the data does not need to be converted from wireless signal to a bit stream.


Has anyone ever noticed something as odd as this? How am I supposed to get the ISP to fix this when they say it's okay as long as wireless connection is "good" enough.

Also, they removed the bandwidth restriction on the testing and the connection is supposed to be capable of 1000 mbit / 1000mbit and they capped out on 450 / 350 with their phone using wireless. Which is fair enough. However they have yet to manage to actually get the speeds the connection should be capable which is 1000 / 1000.

They've meassured 1000 / ~90 with cable and 450 / 350 with wireless.


Does anyone with technical experience has any input or advice here ? Can it be the fiber modem that is not capable of outputing the upload speeds ?


The technicians has also measured the signal on the connection and they say it's fine.
 
Who's routers? If it's their routers, then I would say there is something in the router that is set to cap wired connections. Did anyone test the fiber connection without the router? Plugged straight in it should give 1000/1000 on any device.

Did you do anything to mess with your desktop pcs? I could see you having negative effects if you fiddled around with things like Jumbo frames. With a 9000MTU the downloads would work perfectly fine because you're only going to be getting in packets that are <1500 bytes, but when your pcs send out every one of the packets will be fragmented because they are way too big. Try a live CD on one of your pcs and see if it still happens, that would rule out if you tweaked anything negatively.
 
Yes. The fiber connection has been tested going straight from the modem to the computer and it does not give better upload speed. And like I said we have tested on three different computers. With three different routers, both their brand and Asus Triband and the 68U.

I have not "fiddled" with the network settings or the jumbo packets and even if I did this on one pc. The fact that I have tried three different computers should completely rule this out.
 
I took your advice and ran a live version of ubuntu 7.10 desktop and managed to max out the connection both ways. I have now reseted all network settings back to "default" in Windows 7 and I still don't get the max speeds. In linux I maxed out at 1 gbit / 1 gbit.

Seems that my network settings need tweaking. Any suggestions ?
 
Your networking settings don’t need tweaking, you need an Intel PCI-E NIC.

The Windows Realtek drivers suck.

I see the issue about once a month with chronic speed testers. The crappy drivers are maxing out the CPU or at least one core. Add the stupid HTML5 speed tests into the CPU cycles and it compounds the issue.

Also try downloading the Ookla speed test app in the windows store, it’s a little more accurate on connections 200mbit+.
 
I have looked for that before and there are only server NIC's available.
 
Vaske said:
I have looked for that before and there are only server NIC's available.
Click to expand...

Why is that a problem? PCI-E is PCI-E, doesn’t matter if it has one or 2 network ports. The pro/1000 is one of the most widely supported with the best drivers.
 
Vaske said:
The integrated network card should be equal good to the pro 1000.
Click to expand...

Okay, you aren’t listening. The networks card is fine, the drivers suck. This is confirmed by running Linux and getting good speeds.

Look and see what kind of network card is integrated, I’ll bet it’s a Realtek. Windows Realtek drivers are especially bad.

I deal with this problem so much I carry a pro/1000 with me to show clients the difference and prove that the thruput problem is their device.
 
Does the fiber modem have a WiFi AP? A lot of ISP include a "Their use only" AP with their modem. If so, possible the wifi tests are via their AP.

Is the fiber modem a dumb modem in bridge mode or is it its own router?

Are the test computers connected directly to your router or via a switch?

Might check the packet size on your router and the modem and make sure it matches your Windows PCs. It is supposed to be standard but sometimes ISPs do custom tweaks. If they are using non standard parms, possible the linux setup program is smarter then the Windows one and better matching the connections.

Have you checked for updated drivers? As others have mentioned, a lot of OEM drivers suck for Windows. Don't trust Microsoft to have the latest or best drivers. Check with your MB maker and the maker of the NIC chipset.

As a comparison, on my Win 7 PC, I get about 240mb transfer when copying a large ISO file from my PC to a local HD based NAS. Qualcomm NIC. Path includes about 60' of Cat5 cable and 2 switches. Copying back to the PC's SSD gets about 450mbit transfer.
 
I have found some home routers even though gigabit do not have a back plane that supports such bandwidth. You may want to see what throughput the routers backplane supports, it may explain the behavior you're experiencing.
 
bds1904 said:
Okay, you aren’t listening. The networks card is fine, the drivers suck. This is confirmed by running Linux and getting good speeds.

Look and see what kind of network card is integrated, I’ll bet it’s a Realtek. Windows Realtek drivers are especially bad.

I deal with this problem so much I carry a pro/1000 with me to show clients the difference and prove that the thruput problem is their device.
Click to expand...


Sorry. Didn't mean to come across as ignorant. I didn't explain why. So my motherboard has a intel integrated NIC. It's one of the more "expensive" motherboards. The motherboard is a "Asus ROG Maximus VIII Extreme" running a i219-V integrated NIC


A update on the situation:

Windows 10 seem to give somewhat better uploads but gives worse downloads on speedtest.net testing against various of servers. I am also getting varying results with different browsers.

In Windows 7 I manged to get 850 / 900 with firefox whereas Chrome caps out at around 500 / 92. The same with Opera.

I then tried on my macbook pro. Switching from 2.4 ghz to 5ghz and chrome gave much more desirable results here. 500 down and 800 up. This means that Chrome is not the problem, but rather a compatability with chrome and either the drivers for the NIC or possibly the operating systems where Windows just aren't capable of utilizing a full 1gbit connection.


I have oredered a Intel 1000/Pro as you suggested. Didn't cost me much. I am also considering purchasing a more expensive nic but not sure if it would do much and which one I should go for as they are all targeting server applications and performance.

Either way I will get the PCI-e NIC in less than a week and I shall test more then.
 
Dead Parrot said:
Does the fiber modem have a WiFi AP? A lot of ISP include a "Their use only" AP with their modem. If so, possible the wifi tests are via their AP.

Is the fiber modem a dumb modem in bridge mode or is it its own router?

Are the test computers connected directly to your router or via a switch?

Might check the packet size on your router and the modem and make sure it matches your Windows PCs. It is supposed to be standard but sometimes ISPs do custom tweaks. If they are using non standard parms, possible the linux setup program is smarter then the Windows one and better matching the connections.

Have you checked for updated drivers? As others have mentioned, a lot of OEM drivers suck for Windows. Don't trust Microsoft to have the latest or best drivers. Check with your MB maker and the maker of the NIC chipset.

As a comparison, on my Win 7 PC, I get about 240mb transfer when copying a large ISO file from my PC to a local HD based NAS. Qualcomm NIC. Path includes about 60' of Cat5 cable and 2 switches. Copying back to the PC's SSD gets about 450mbit transfer.
Click to expand...


The Fiber modem does not have a AP. It's just a modem, not a router. I think I have established that it's not the modem or the connection, well, for the most part. I also believe the ISP might have done something because I wasn't even getting these results in Linux some weeks ago. I guess all my complaining to the ISP worked if that's the case.

The problem seem to lie in utilizing the bandwidth.


I might also add that the connection does seem to produce less desirable results during peak hours which I find odd seing how it's supposed to be a dedicated connection.
 
bds1904 said:
Your networking settings don’t need tweaking, you need an Intel PCI-E NIC.

The Windows Realtek drivers suck.

I see the issue about once a month with chronic speed testers. The crappy drivers are maxing out the CPU or at least one core. Add the stupid HTML5 speed tests into the CPU cycles and it compounds the issue.

Also try downloading the Ookla speed test app in the windows store, it’s a little more accurate on connections 200mbit+.
Click to expand...


I managed to cap out at 108 mb / sec downloading something trivial from steam. However with torrents i'm rarely getting 20 mb / sec either direction.
 
Vaske said:
Sorry. Didn't mean to come across as ignorant. I didn't explain why. So my motherboard has a intel integrated NIC. It's one of the more "expensive" motherboards. The motherboard is a "Asus ROG Maximus VIII Extreme" running a i219-V integrated NIC


A update on the situation:

Windows 10 seem to give somewhat better uploads but gives worse downloads on speedtest.net testing against various of servers. I am also getting varying results with different browsers.

In Windows 7 I manged to get 850 / 900 with firefox whereas Chrome caps out at around 500 / 92. The same with Opera.

I then tried on my macbook pro. Switching from 2.4 ghz to 5ghz and chrome gave much more desirable results here. 500 down and 800 up. This means that Chrome is not the problem, but rather a compatability with chrome and either the drivers for the NIC or possibly the operating systems where Windows just aren't capable of utilizing a full 1gbit connection.


I have oredered a Intel 1000/Pro as you suggested. Didn't cost me much. I am also considering purchasing a more expensive nic but not sure if it would do much and which one I should go for as they are all targeting server applications and performance.

Either way I will get the PCI-e NIC in less than a week and I shall test more then.
Click to expand...

Chrome and Opera are based on the same underlying code. So I think we're getting closer to the problem here. Do you have any plugins enabled? Maybe try doing a speedtest in Incogneto mode and see if it works as expected. The fact that Firefox delivered the rated speeds means that everything is working just fine, so it's just coming down to a browser problem.

The Pro 1000 while a great card, is really, really old. It has been depreciated in some OSes so a more modern variant like an i210 would be appropriate. The main issue is that the Pro-1000 is like the Artic Silver of the network world. It's the most well known so people will keep recommending it regardless if there are newer, better options. I don't think there is going to be anything wrong with your onboard NIC that you have either.

Chrome on Mac isn't the same on Chrome on Windows, so I wouldn't try to compare the two. If you're getting ~500+ on Wifi, that's about all the more you should expect anyway, so I'd say that part is working just fine.

The one thing I will agree on is that I wouldn't put too much stock into a browser speed test for testing 1gig. At a certain point it's just not possible to make a single connection saturate a network link, and 1 gig is a ton of bandwidth to try to do so. You likely are not going to see everything capable of giving you max speeds every time you try to do 1 task. But if you have 5 - 10 devices doing many things at once, the overall throughput should be greatly improved. As for the torrents, that one is going to be really convoluted so it could come down to 100 different reasons why those perform poorly.
 
This thread is super confusing to follow. You have your fiber coming in to an ont? From here you have Ethernet connection coming from the ont into your own router/wifi combo? Can you do a test with a straight with a single computer wired directly to the ont?
 
Vengance_01 said:
This thread is super confusing to follow. You have your fiber coming in to an ont? From here you have Ethernet connection coming from the ont into your own router/wifi combo? Can you do a test with a straight with a single computer wired directly to the ont?
Click to expand...

You should read the replies before posting.
 
bman212121 said:
Chrome and Opera are based on the same underlying code. So I think we're getting closer to the problem here. Do you have any plugins enabled? Maybe try doing a speedtest in Incogneto mode and see if it works as expected. The fact that Firefox delivered the rated speeds means that everything is working just fine, so it's just coming down to a browser problem.

The Pro 1000 while a great card, is really, really old. It has been depreciated in some OSes so a more modern variant like an i210 would be appropriate. The main issue is that the Pro-1000 is like the Artic Silver of the network world. It's the most well known so people will keep recommending it regardless if there are newer, better options. I don't think there is going to be anything wrong with your onboard NIC that you have either.

Chrome on Mac isn't the same on Chrome on Windows, so I wouldn't try to compare the two. If you're getting ~500+ on Wifi, that's about all the more you should expect anyway, so I'd say that part is working just fine.

The one thing I will agree on is that I wouldn't put too much stock into a browser speed test for testing 1gig. At a certain point it's just not possible to make a single connection saturate a network link, and 1 gig is a ton of bandwidth to try to do so. You likely are not going to see everything capable of giving you max speeds every time you try to do 1 task. But if you have 5 - 10 devices doing many things at once, the overall throughput should be greatly improved. As for the torrents, that one is going to be really convoluted so it could come down to 100 different reasons why those perform poorly.
Click to expand...


Well. I have a 219-V. Is the 210 better? I tried disabling all the extensions in private mode (opera). Same results.
 
Vaske said:
Well. I have a 219-V. Is the 210 better? I tried disabling all the extensions in private mode (opera). Same results.
Click to expand...

Probably not. When the product numbers are that close chances are good you just have a variant of the same chip. It's not like they are really innovating in that marketspace and it's just more of a standard feature set. The 219-V is about 3 years newer, and has a smaller die. If I were guessing the 219 is just another variant of the 210 that is on a newer process so it uses less silicon, power, and costs less to produce. Both of them are expected to be discontinued around 10 years into the future, so support should be around for a while.

https://ark.intel.com/compare/64402,82186

As far as why that browser is still giving you grief, I'd say I'm at a loss at this point. It is weird that the browser is only giving you <100mbits of traffic, as there is no reason why an modern pc couldn't handle that even in software. I remember only being able to hit ~150mbps using 100% cpu on a stripped down single core VIA 1.4ghz processor, but even a P4 has no real issues pulling >100mbps. But I guess re-reading the thread you didn't say what site you were using? Speedtest.net should give fairly accurate results. But at this point I'm wondering if your provider doesn't have something that is just detecting speed tests and deprioritizing them or just messing with the traffic in general. It might pick up certain traffic flows but not others. A real test is a sustained upload somewhere, but even things like SFTP cap out around ~200mbps so once again it would be hard to really test a gig unless you had multiple tests going. I get the fact that you're concerned that you're not getting the speeds you are paying for (which is only 250 / 250 in your case) but I wouldn't put stock in just a speed test alone to tell you if you're getting it. Upload something large like a Windows 10 ISO to Google Drive or Onedrive and see what happens. It's possible one of those might go faster than the speedtest does.
 
219V has no problems maxing gigabit, I have quite a few and they all hit real world 940Mb TCP. Anything intel supported by their mainstream drivers is known to be good for line rate on any normal computer, even stuff that has been discontinued for awhile. 82579: chipset of champions.

The server nics are nice but not faster on the same interfaces (no consumer 10/40 models from intel yet...aquantia has a good 10GbE though) but they also support more high end features such as SR-IOV, advanced offload etc.

Sounds like its either cables, router or software being stupid.
 
Like I wrote in the OP. All the cables have been swapped out for new ones. Used to have cat 6, now they are all cat 7 with shielding. Also the routerh as been swapped out for another higher model of the same brand (Asus). I've also tried a different brand being a third router. Same results. So having in mind i've changed out the computers, cables and routers, what else could it be ?
 
I mean you've already proved that you can get gigabit in some of your configurations. You maxed out the connection using Firefox, and you also hit the limits using a live cd, and your Mac also likely hit the limit. So other than the feel good of seeing pretty numbers, are you actually noticing issues?

But long story short, we don't have a clue how many pcs are in play, what configurations you've tried, what site your trying to go to, what time of day you're attempting to benchmark, and what other stuff is running on the network. There's a pretty good chance if you run any type of torrent software, the open connections from that will just obliterate any soho device, and it's never really going to be possible to max out a gigabit connection using torrents. Your ISP might have an IPS system which detects that torrents are coming from your connection and throttling your traffic. The ISP might also have bandwidth issues on their end and certain paths from their network might have slower speeds. I can't say for sure from this side of the fence if each machine you're running a speed test from is even hitting the same speed test server, which could provide wildly varying results. Time of day can also have an impact.

Test with 1 and only 1 system connected to the network at like 1am, and make sure it's not the one with the torrent software on it. If the box is reporting zero traffic before doing a speed test and you still only get a slow speed in a certain browser, then you need to figure out why that browser sucks. I think it was already mentioned that you might have say an HTML5 test in something like chrome, but it could be a flash test in IE, or something even different still in firefox. And depending upon the test, none of those tests actually last long enough to allow some systems to ramp up. A 100MB speed test will only take 1 second to complete, where the tcp window might take 15 - 20 seconds to really get up to speed if it starts out slowly. A speed test will be useless to really show you that. But a download from Steam should have plenty of time to pick up speed.
 
I don't think that CAT6 or CAT7 are needed. I would try it without the router and plug your desktop directly into the modern. The Intel 219 is fine. I would also test speeds between two wired devices on the network to check if the router can handle that too...
 
You must log in or register to reply here.
Back
Top