Riddle Me This, Nighthawk!

[Revanchist1ab]

Router: Netgear Nighthawk X6S AC3000 (R7900P)

 

Scenario:

About 20+ connected devices:

• 3 PCs (2 of which are usually turned off)
• 3 phones
• 2 iPads
• 4 Kindles
• 4 Alexas
• 4? Google Home Minis

I have gigabit Ethernet from AT&T and I get an average of 950 down on my AT&T router/modem which works in bridge mode to connect to my Netgear which gets about 850 to 900 (cause Uverse connects to modem and that sucks some up I guess... whatever).

 

Problem:

My gaming PC which is used primarily at night can only get 400 down on average. Every so often, it can max out at 600, but that's just an initial jump during a test. It's 400, 400, 400. It can get higher, up to 900+, but that was only when nothing else was connected. So, the question is whether I can get up to a sweet 700. At night, the phones are not in use, the other PCs are turned off, and the iPads and Kindles are usually either off or in sleep mode. The smart home devices are anyone's guess. I personally hate the things, but family enjoys them.

 

I've played with the router ad nauseam, fiddling with settings, even the QoS which everyone tells you to shut off, but it doesn't change the outcome of forever 400. My CCNA and CySA+ training aren't helping against the onslaught of (if the Internet is correct) utterly bad software.

 

I am well aware I don't NEED 700, but I WANT it. I am also well aware everyone hates this router... I got it from Costco during Christmas and I'm going to be replacing it this year, but I do want to know if this can be fixed before that... for funsies.

 

Final note: God, I hate this Netgear software. The "Attached Devices" section is broken as not only does it not show the download and upload speeds, but it doesn't save the icons you can select for each device. Like... I know its nitpicky, but if it can save the device priority, why not the icon? It may be minor but WHY? Who forgot to code this!!! (P.S. Other people have had the monitoring issues and its apparently not limited to edge cases, so I don't care to fix it.)

[KarathKasun]

WiFi bandwidth is generally only 1/2 of the WiFi connection speed.  Which means 650 is the absolute max with perfect signal if you actually get a 1300 connection.  Most WiFi adapters seem to max out at 867.

 

This is on ANY WiFi router.  WiFi6 is supposed to address this somewhat, but its pretty expensive to implement.

[Revanchist1ab]

@KarathKasun Not on WiFi my dude. Physical connection AND I disabled WiFi on my PC.

 

Running CAT7, btw.

[moriel5]

Apparently, it looks like the R7900P is a watered-down version of the R8000P, from what I can gather on WikiDevi, Reddit, and the Netgear and DD-Wrt forums. It may be possible to run on it the R8000P DD-Wrt firmware, however that would obviously nullify the warranty.

 

How about getting the LinkSyS Wrt3200ACM instead, since there is nothing better on the market (it doesn't support WiFi 6/Wireless-AX, however that is it's only shortcoming), especially since it LinkSyS endorse flashing OpenWrt on it (you can even dual-boot on it)?

[Revanchist1ab]

@moriel5 Yeah, I don't want to try anything that drastic, yet. I was just curious if anyone knew what was causing the router to limit my PC's throughput.

 

Oh, and I saw that Linksys, but people are complaining about quality control, due to "a bad chipset." Regardless, not great.

[moriel5]

1 minute ago, Revanchist1ab said:

@moriel5 Yeah, I don't want to try anything that drastic, yet. I was just curious if anyone knew what was causing the router to limit my PC's throughput.

Did you check with another PC? Perhaps something else was limiting it?

[Revanchist1ab]

@moriel5 K... So just ran a basic network speed test (10 PM on a Monday EST) through Ookla using the same CAT7 patch cord on my laptop and PC.

 

Laptop: Acer Aspire E17 (E5-774G-52W1)

PC: Acer Predator AG3-710-UC11 (DG.B14AA.008)

 

My laptop which has a cleaner setup and has a newer NIC gets an average of 580 with a cap of 700.

My PC which is a year and a half older, gets an average of 465 with a cap of 600 in the beginning.

 

Both are running Norton AV, but my PC is also running Malwarebytes and has more software installed. I can almost assuredly bet that something is stealing that extra 100, but I doubt its that much. The initial caps are great, but I'd like to see that be at least 100-200 higher as these speeds are just during quiet hours on a weekday.

[moriel5]

29 minutes ago, Revanchist1ab said:

@moriel5 K... So just ran a basic network speed test (10 PM on a Monday EST) through Ookla using the same CAT7 patch cord on my laptop and PC.

 

Laptop: Acer Aspire E17 (E5-774G-52W1)

PC: Acer Predator AG3-710-UC11 (DG.B14AA.008)

 

My laptop which has a cleaner setup and has a newer NIC gets an average of 580 with a cap of 700.

My PC which is a year and a half older, gets an average of 465 with a cap of 600 in the beginning.

 

Both are running Norton AV, but my PC is also running Malwarebytes and has more software installed. I can almost assuredly bet that something is stealing that extra 100, but I doubt its that much. The initial caps are great, but I'd like to see that be at least 100-200 higher as these speeds are just during quiet hours on a weekday.

Hmm... Perhaps Windows Update in the background? Or perhaps the Microsoft Store is updating apps in the background?

[Revanchist1ab]

Unless its another hidden setting, I have had Windows auto-updates turned off since I turned the computer on. While I completely doubt Microsoft store is updating the 2 store apps I have, is there a setting to disables that?

[moriel5]

6 hours ago, Revanchist1ab said:

Unless its another hidden setting, I have had Windows auto-updates turned off since I turned the computer on. While I completely doubt Microsoft store is updating the 2 store apps I have, is there a setting to disables that?

In the Store's settings you have an option to enable auto-updates, an option that is enabled by default.

And I find it hard to believe that you only have 2 store apps installed, that only possible on the LTSC edition, since it doesn't come with any store apps at all (you install the store along with the most basic dependencies, unofficially).

[KarathKasun]

19 hours ago, Revanchist1ab said:

@KarathKasun Not on WiFi my dude. Physical connection AND I disabled WiFi on my PC.

 

Running CAT7, btw.

What NIC is used in the system, Intel or Realtek?

 

There are some network adapter settings you can tune on the PC itself like jumbo frame size, interrupt moderation, buffer sizes, etc.  Seeing that you have a system getting higher results, the problem is likely with the PC configuration and not the router.

 

You can get to these settings through device manager, they are on the properties page for the NIC.  For better performance at a small CPU overhead cost, disable interrupt moderation.  Set Jumbo frame size to 9k.  Leave buffers alone for now.

[moriel5]

2 hours ago, KarathKasun said:

What NIC is used in the system, Intel or Realtek?

 

There are some network adapter settings you can tune on the PC itself like jumbo frame size, interrupt moderation, buffer sizes, etc.  Seeing that you have a system getting higher results, the problem is likely with the PC configuration and not the router.

 

You can get to these settings through device manager, they are on the properties page for the NIC.  For better performance at a small CPU overhead cost, disable interrupt moderation.  Set Jumbo frame size to 9k.  Leave buffers alone for now.

Yeah, that's a valid point.

Realtek's networking products are pretty bad due to incompetent drivers (this I have heard from people working on the Linux Realtek drivers, I was a beta tester for a short period of the drivers for the RTL8211AE (WiFi-AC)).

 

Other than that, while jumbo frames do make more efficient usage of the network, and as such improve networking speed, it won't help speed up the Internet speed, since to my knowledge, no ISP makes use of jumbo frames for non-corporate clients, and even then, only a small percentage actually use it, such as universities, some mega-corporations, and some governmental uses.

 

There should be no need to touch buffers either way.

[mynameisjuan]

2 hours ago, moriel5 said:

Other than that, while jumbo frames do make more efficient usage of the network,

Unless you know what you are doing they do not make more efficient use of the network. They have a very specific use case.

 

Never use jumbo frames. Fragmentation will hurt not only your internet performance but also your router most likely cannot handle it.

[moriel5]

1 hour ago, mynameisjuan said:

Unless you know what you are doing they do not make more efficient use of the network. They have a very specific use case.

 

Never use jumbo frames. Fragmentation will hurt not only your internet performance but also your router most likely cannot handle it.

I will need to read upon the matter before I can say yay or nay to your comment, however from my experience, I have yet to see fragmentation as a result of jumbo frames, rather than because lack of jumbo frames.

[mynameisjuan]

5 minutes ago, moriel5 said:

I will need to read upon the matter before I can say yay or nay to your comment, however from my experience, I have yet to see fragmentation as a result of jumbo frames, rather than because lack of jumbo frames.

Fragmentation is a direct result of jumbo frames. The moment a jumbo frame hits an interface less than its MTU it’s fragmented unless it’s DF is 1 then dropped

[moriel5]

32 minutes ago, mynameisjuan said:

Fragmentation is a direct result of jumbo frames. The moment a jumbo frame hits an interface less than its MTU it’s fragmented unless it’s DF is 1 then dropped

That certainly makes sense, and also explains how I saw the opposite in my usage.

It depends on what kind of usage it is.

[mynameisjuan]

12 minutes ago, moriel5 said:

That certainly makes sense, and also explains how I saw the opposite in my usage.

It depends on what kind of usage it is.

Im just not sure what usage you are referring to but I can only guess is you mean it takes multiple packets to send the same amount of data a jumbo frame can fit. That is NOT fragmentation. 

 

Fragmentation only occurs when the original packets has a larger MTU than any device in its path. If you dont enable jumbo frames on the end device, 99% of the time there will be no fragmentation. 

 

Again, this should only be used on end to end circuits

[moriel5]

1 minute ago, mynameisjuan said:

Im just not sure what usage you are referring to but I can only guess is you mean it takes multiple packets to send the same amount of data a jumbo frame can fit. That is NOT fragmentation. 

 

Fragmentation only occurs when the original packets has a larger MTU than any device in its path. If you dont enable jumbo frames on the end device, 99% of the time there will be no fragmentation. 

 

Again, this should only be used on end to end circuits

Thanks. In that case, I really need to read up more, since so far I have benefited from jumbo frames (admittedly, only up to ~4.5KB), however I only use it locally, and I don't utilize Windows for this either, however it could be placebo.

[Revanchist1ab]

@KarathKasun It's a Realtek PCIe NIC on an Acer Predator pre-built circuit board.

https://snlookup.com/acer-predator-g3-710-desktop-dg-b14aa-008-p7197#ffs-tabbed-16

 

I have about as much faith in it as I do a bridge in the woods made by a middle-school youth group.

 

I'm at the point where I'm going to stop caring soon as I doubt there's anything that can be done beyond replacing both the router (new one comes in two days) and the PC (gonna be swapped out in a year... if I can find a job).

[moriel5]

12 minutes ago, Revanchist1ab said:

@KarathKasun It's a Realtek PCIe NIC on an Acer Predator pre-built circuit board.

https://snlookup.com/acer-predator-g3-710-desktop-dg-b14aa-008-p7197#ffs-tabbed-16

 

I have about as much faith in it as I do a bridge in the woods made by a middle-school youth group.

 

I'm at the point where I'm going to stop caring soon as I doubt there's anything that can be done beyond replacing both the router (new one comes in two days) and the PC (gonna be swapped out in a year... if I can find a job).

That looks pretty good, except for the Realtek NIC, which is not surprising, given that 99% of the motherboards utilize Realtek NICs, since that is the cheapest option.

I'm pretty confident that you could leave things be on the computer, since you cannot replace the NIC (you could buy a Gigabit PCIe NIC, however those are prohibitively expensive for no reason, unless it is Realtek, in which case you get pretty much the same quality whether it is a ~5$ PCIe NIC or a ~90$ PCIe NIC, and it will be the same as the onboard NIC).

 

One more word about Realtek networking driver flaws:

Unlike other manufacturers' NICs, Realtek's NICs do not utilize hardware processing, only software processing, which is why the network speed will be heavily dependent on the system's overall performance (which is why Realtek performs alright on high end devices (while taking away a bit of the system's performance), but much worse on low-end devices (and for laptops, cause a small, but additional battery drain).

[Revanchist1ab]

@moriel5 @mynameisjuan Just to conclude this topic, I plugged in a new router, did some testing back and forth and I'm sure its this computer.

 

I haven't a clue why my connection is so slow as I had it up to 900 ages ago, but it is what it is.

 

Current connection is running at 650 down max, 425 down average with 360 up.

Disabling both antivirus suites and firewalls nets a negligible bandwidth increase.

Network test uses about 50% of the CPU.

 

Who knows. Maybe its all the cheap components as this IS a manufacturer pre-built.

[KarathKasun]

On 9/13/2019 at 1:19 PM, Revanchist1ab said:

@moriel5 @mynameisjuan Just to conclude this topic, I plugged in a new router, did some testing back and forth and I'm sure its this computer.

 

I haven't a clue why my connection is so slow as I had it up to 900 ages ago, but it is what it is.

 

Current connection is running at 650 down max, 425 down average with 360 up.

Disabling both antivirus suites and firewalls nets a negligible bandwidth increase.

Network test uses about 50% of the CPU.

 

Who knows. Maybe its all the cheap components as this IS a manufacturer pre-built.

Tune the NIC settings.  Its possible a driver update was pushed that has worse default tuning options.

[Revanchist1ab]

@KarathKasun Unfortunately, I haven't the foggiest of which specific settings I should change.

[KarathKasun]

1 hour ago, Revanchist1ab said:

@KarathKasun Unfortunately, I haven't the foggiest of which specific settings I should change.

Turn on Jumbo frames and all that.  Literally just fiddle with it till it works properly in your network.  Change a setting, see how it reacts.

 

Jumbo frames should reduce the CPU usage.

 

Maybe screenshot the defaults first, so you know for sure what the defaults are.