Decreased network performance after hardware upgrade

[Proxipupuce]

Yeah, that's a head scratcher. You should never see local speeds < internet speeds. Usually your internet speeds will be less unless you're one of those lucky ones that have a 1G internet connection. Mine, for example is only 350 Mbps max and that's only during off-peak times. Realistically, my internet speed is more like 300M.

I tried other things.
- Reconfigure my router perfectly
- Uninstall and reinstall the network driver of my desktop

While playing with speeds and duplexes I realize something:

If I set the speed to 100 Mbps full duplex, I block at 92 Mbps or 10 MB/s (local and internet)

On the other hand, if I leave at 1Gbps full duplex, I stay at 160 mbps in local.

So I am in 1 Gbps mode, otherwise I would be stuck at 92.

I really don't understand where this bug comes from...

 

[Proxipupuce]

Also, when I install the new network card in the server, I will have to install drivers?

 

[Whattteva]

Also, when I install the new network card in the server, I will have to install drivers?

No, should already be built-in and work out of the box.

 

[Proxipupuce]

No, should already be built-in and work out of the box.

Well, I just have to wait for the new network card. I'm a bit disappointed not to be able to take the intel i210, but I think the Intel Gigabit CT Desktop Adapter will be very good, at least better than my Realtek.

I'll let you know this week after the installation!

Thank you very much for spending the day helping me, I would never have made any progress alone!

 

[Proxipupuce]

Hello everyone, I bring news about my problem.

I received today an Intel I210 T1 board modified by HP. I installed it, and apart from a clear improvement in stability and a slight increase in throughput, my problem is still there.

I disabled the Realtek controller built into the motherboard, and I'm now sure that my new card only works at 1Gbps Full-Duplex.

After trying it on many different computers, both ethernet and wifi, none of them exceed 120mbps with an iperf test, even the one that was working before.

In samba transfer, I reach 200mbps, and in internet download 250mbps.

So I wonder if my network settings are good:

DNS: 8.8.8.8 / 1.1.1.1
MTU : I was thinking of setting it to 9000. My network cards and router support Jumbo Frame. Is this a good idea?
Hardware unloading: I don't know if I should activate it?

 

[Proxipupuce]

If anyone has any idea where my problem might be coming from, a poorly set parameter or whatever, I'd gladly accept help.

 

[Whattteva]

Maybe the CPU single-thread is too weak? You can try running iperf3 in parallel on the client.

Code:

iperf3 -P 2 -c <server_ip>

You can play around with that number after -P (2, 3, or 4) and see what results it gives you.

 

[Proxipupuce]

Maybe the CPU single-thread is too weak?

What do you mean by "single thread"? The i5 3570, despite the absence of Hyper-threading, has 4 threads, like the i3 I had before. The only difference was that the i3 had only two cores and used Hyper-threading.
Otherwise, I still have the i3, I could try installing it to see...
When in doubt, I reinstalled truenas, which allowed me to switch to UEFI.
I'll try out the command later, on my way home.
Thanks a lot!

 

[jgreco]

1Gbps is 900 Mb/s, not 900MB/s, right?

That's correct, and a good reason not to cheap out on the "yte" or "it" keystrokes.

1MByte/s is ideal, but everyone will consider it equal to 1mbyte/s even without the caps.

People might look at 1MBIT/s a bit strange but it is still clearly 1Mbit/s.

 

[Whattteva]

That's correct, and a good reason not to cheap out on the "yte" or "it" keystrokes.

1MByte/s is ideal, but everyone will consider it equal to 1mbyte/s even without the caps.

People might look at 1MBIT/s a bit strange but it is still clearly 1Mbit/s.

I like to use the term MebiByte (MiB) or GibiByte (GiB) to further disambiguate between M (10E6) vs Mi (1024E3).

 

[Whattteva]

What do you mean by "single thread"? The i5 3570, despite the absence of Hyper-threading, has 4 threads, like the i3 I had before. The only difference was that the i3 had only two cores and used Hyper-threading.

I'm not saying the CPU itself is single thread. I'm saying the iperf test may have to be run with that -P flag to make it more multi-threaded.

 

[Proxipupuce]

Test results :
-c:163 Mbits/s
-P 2 -c: 160 Mbits/s
-P 3 -c: 163 Mbits/s
-p 4 -c : 166 Mbits/s

I got these results with the old laptop, which didn't go beyond 100 Mbits/s before. Same thing with samba sharing, before not above 20 Mbytes/s, now 35 easily. All this on wifi.
I rebooted this computer, and now I have these results:

-c: 64 Mbits/s
-P 2 -c: 120 Mbits/s
-p 3 -c: 140 Mbits/s
-p 4 -c: 153 Mbits/s

Samba: 25 Mbytes/s, as before.

With another wifi laptop, the results are similar.
I'm trying with my desktops at 9pm.

 

[Proxipupuce]

I'm not saying the CPU itself is single thread. I'm saying the iperf test may have to be run with that -P flag to make it more multi-threaded.

My bad, I didn't understand!

 

[Proxipupuce]

After several tests, I have exactly the same result on my desktop PC. The only notable thing is that when I connect a device to the same repeater as my NAS, the network works perfectly at 980 Mbits/s. I assume that the repeater directly connects the devices?

 

[Whattteva]

After several tests, I have exactly the same result on my desktop PC. The only notable thing is that when I connect a device to the same repeater as my NAS, the network works perfectly at 980 Mbits/s. I assume that the repeater directly connects the devices?

Are you talking about WiFi repeater? Is it using an Ethernet backhaul or pure wireless? Cause your effective bandwidth will be halved if that's the case. Though, I think if all devices that need to talk each other are on the same repeater, it should not incur that overhead. It's only if it needs to hop back to the router.

 

[Proxipupuce]

I use a repeater network in my house. For some reason, the wifi on my box is extremely bad. And since my house isn't wired for Ethernet, I'm dependent on wifi. So I have three ORBI repeaters in my house. One is Ethernet-connected to the box and broadcasts wifi, and the others repeat on the first and top floors. All the repeaters form a single network. I have Ethernet ports on these repeaters, to which my NAS is connected.
My Wi-Fi laptops connect to these repeaters. For my desktop, it's a bit stranger: it's connected via Ethernet to an ASUS RT-AX 55 router, which picks up the Wi-Fi signal from the ORBI repeaters (I know this isn't optimal).

 

[Whattteva]

I use a repeater network in my house. For some reason, the wifi on my box is extremely bad. And since my house isn't wired for Ethernet, I'm dependent on wifi. So I have three ORBI repeaters in my house. One is Ethernet-connected to the box and broadcasts wifi, and the others repeat on the first and top floors. All the repeaters form a single network. I have Ethernet ports on these repeaters, to which my NAS is connected.
My Wi-Fi laptops connect to these repeaters. For my desktop, it's a bit stranger: it's connected via Ethernet to an ASUS RT-AX 55 router, which picks up the Wi-Fi signal from the ORBI repeaters (I know this isn't optimal).

Ooohhh. Now your crappy and inconsistent throughput rates make sense. It's your wifi repeaters lol. Even if you connect through ethernet to the repeater itself. It's kinda useless since you're essentially limited by the wifi back haul, which is gonna be shit. Even if you outfit your car with the best quality Lamborghini skin, tires, etc from the year 3000 It's still gonna be shit if the engine is a geo from 1980s lol.

 

[Proxipupuce]

I'm back with more news!
Today, having two Ethernet ports available on my NAS, I decided to do some network aggregation.
So I went into the BIOS, reactivated the integrated controller and restarted the server.
And since then, I don't know why, but I haven't had any network problems. Since there are two interfaces, I've been saturating my wifi (at 35 MBytes/s) on all my devices, even my desktop.
So I set up a FAILOVER aggregation as I wanted to do, with the motherboard port plugged into my old PLC and the new card plugged into the repeater, and everything works perfectly.
Throughput is stable, from 30 to 35 MBytes/s on all my devices.
Before, I had trouble getting over 20 and it wasn't stable.
Does anyone have an explanation?

 

[jgreco]

Does anyone have an explanation?

Sure. Because you have a crappy network topology. Things work poorly as a result. You can make random changes and induce different optimal paths that perform differently. Sometimes better, sometimes worse.

 

[Proxipupuce]

I'm aware that my network is very poor. The problem is that I have an old house, with a box at the bottom and dozens of devices to connect on two floors, which sometimes don't support Wi-Fi.
Repeaters seem to me to be the best option, given that I can't have a wired house, or 40 devices directly connected to the box.
In any case, thank you all for your time.