it is very well possible you’re hitting your CPUs maximum capacity for handling data. that’s why you typically run iperf in parallel mode (-P flag). also it’s worth noting, that tcp (default) adds overhead, thus testing with udp (-u) might be useful too.
You probably want to tune the network quite a bit when it comes to multi-gigabits network. There is tons of configurable params with sysctl (like buffer sizes) and ethtool (lro, integrity check) that will impact network speed. I.e. I have a 10Gbps link between my firewall and my desktop, with or without tuning lead to very big difference, like without tuning, it could be stuck at ~4Gbps, with tuning, I was able to get up to nearly full 10Gbps. Obviously it could be case by case, and could very well be USB chip throttling. However, without tuning the system network params, I think there is also a good chance that the network speed was limited by the OS. Most OS are optimised for 1Gbps and expect multi-gigabit network owner to have the knowledge to perform tuning themselves.
thanks for watching mate and your comments! :) I have 3 networks in my house! 1 x gig and 2 x multigig (2.5 and 10) Although the 10 one is peer to peer with my NAS as I don't need it for anything else. But mate, none of them gave me half the hassle to configure as this device! Big pain in the but! I think it's mostly due to drivers (maybe the device itself)
@@DigitalMirrorComputing I think some OS could be more optimised for 10G, and some requires more configuring. I am running a OPNsense with CX342, and my desktop running Arch Linux (and Windows) with HP 560SFP+ (Intel 82599ES chip). I found that I run 10G smoothly with Windows, minimal configuration, likely only changing MTU. But require more tuning on Linux. For the 10G one you have, if you are running Windows to a NAS directly running 10G, probably will be quite easy, as I think both the NAS OS or off the shelf NAS will likely optimise more for 10G as well.
@@DigitalMirrorComputing i have experience with 2.5, 10, 25 and 40gbit. But not with 5. But since 2.5, 5 and 10gbit work exactly the same just with divided signal rate i think the same rules apply. So not reaching full speed gets more of a problem when speed inreases, things like actual buffer space in the hardware and offloading capabilities of the used network chipset on the card become more important. Also the way you connnect the card becomes more important. Although USB is fast enough there is more CPU overhead so on low end devices the CPU might become a bottleneck if you combine it with a USB connected card. Also did you test with just normal frame size of just 1500 because that is the standard every piece of hardware supports in the correct way. Jumbo frames upto 9000 and Super jumbo frames should only be used if you are 100% sure your complete route supports it. If not sure just don’t use it as it can cause all kind of problems in for example vpn software. And you can reach about 4.6 to 4.7gbit with a mtu of 1500 anyway so why bother tweaking with jumbo frames. A mtu of 1500 makes you loose about 3.6% of the theoretical speed with jumbo frames (mtu 9000) you reduce that to 0.6% i refuse to even calculate it for Super Jumbo Frames as like 99% of consumer hardware does not support it correctly anyway. Even in enterprise gear it is not that common.
TrueNas is know for not supporting 5gb and alot of 2.5gb addin cards. So I dropped a 10gb sfp+ dual port nic in my home lab as they are fairly affordable and enuse I has the highest throughput it can for my 2.5gb home network. It might seem like a bit of overkill.....and it is. But it also works beautifully.
didi you put the usb-c stick in at least a 10 gigabit socket. the overhead from the usb protocol might be slowing your transfer speeds down if you are using a 5 gigabit socket
I tried several mate! 10Gb and 20Gb USB-C. Same results! also used an adapter on 2 different computer USB-A 3.0 (with limit of 5Gb) same results unfortunately!
I have used 1 Gbps USB adapters for years on my laptop which doesn't have Ethernet. Works perfectly fine. However each and every 2.5 Gbps USB adapter I have tried ranged from "occasionally unstable" to "just unusable", even when running on 1 Gbps link. Those 2.5 Gbps USB ones I tried which were "occasionally unstable", could run stable for hours and sometimes even days but suddenly refused to do anything unless you disabled/enabled the ethernet device or replugged it. It claimed to be connected, but no traffic possible. This was about 3 - 4 years ago, so I was hoping they may have fixed those issues as they also were happening with those onboard Realtek 2.5 Gbps adapters. I clicked your video in the hope to see a positive experience from you with these. For example on a laptop and also for longer periods (e.g. on desktop/server PC) But given your experience with them, I guess these have still the same kinds of issues as those 2.5 Gbps ones years ago.
My feelings exactly mate! I saw quite a few people talking about these dongles online and giving them stellar reviews but unfortunately that was not my experience! Maybe it was just me, but I feel that the drivers and kernel support just isn't there IMO at least! Maybe you can get better results if you fiddle more with it, but again, IMO that is not a good user experience and hence we can't categorize it as a "GREAT" product. MAx I could get out of it was around 3Gb. And like you point out, on the laptop it was a struggle as well at first!
CAT 7 cable is not what you used because it is a proprietary format and these devices have a normal RJ45 connectors. You probably have been sold a shieled cat 6A that the seller advertised as CAT 7. It could also be a CAT 8 cable but from the seller perspective it would be stupid to under advertised what you sell. Also CAT 8 cables are a pain to use because they are not flexible at all. Also CAT 6 (up to 55m) and CAT 6A (up to 100m) are capable of 10Gbit.
On my win pc (personal computer) I used a USB-C 10Gb, on the pi I used USB-A which should give you 5Gb. but in neither of those I managed to go over 3Gb! Hope that helps!
@@DigitalMirrorComputing It's possible that the USB-C cable is soldered halfway and to get full speed you need to use the USB 3.2 Gen 2 standard to avoid a bottleneck in the USB connection, although this is unlikely, it's worth checking. Thanks for the video.
@@Иван_Абрамов could als be that the USB does add a bit more in latency which limits the transfer rate. If so, then multiple parallel streams may increase total used bandwidth. Can be tested with iperf for example
To get 5gbe over usb you need to plug into a 10gb USB port, if you plug it into a 5gb USB port the most you’ll get is about 3.5gb, there’s always a overhead.
8:12 Its not a "hit or miss" lol. Its skill issue or in this case no dedicated equipment to route (router). By doing direct connection, you're essentially telling both ends to route which would be fine if it wasn't USB. Getting a multi-gig router or diy pfsense would solve your problems.
well, it sounds like you are a lot smarter than me, so I would say go ahead and get those devices! :) However I must add (for the people reading this) that this comment lacks in (not just courtesy) but accuracy. In a peer-to-peer connection (aka direct link) no router is needed! The two devices can communicate directly with eachother provided (and they are) on the same network! I also don't understand the point around USB... When two devices are connected directly using network adapters (even if they are USB), each device’s network stack handles the direct communication. Routing typically refers to the process of directing traffic between different networks (like from your local network to the internet), and a direct connection does not involve that kind of routing. There are no routing tables involved in a peer-to-peer setup on a single subnet. Routers route traffic! Hope that helps!
![COMPLETELY Overhauling This Old HP Server [Part 1]](http://i.ytimg.com/vi/2HLF-46Das4/mqdefault.jpg)
the "pure chinesium" rule should apply. It will work 60% of the time on 40 % of the stated capacity/speed/power rating etc.
it is very well possible you’re hitting your CPUs maximum capacity for handling data. that’s why you typically run iperf in parallel mode (-P flag). also it’s worth noting, that tcp (default) adds overhead, thus testing with udp (-u) might be useful too.
You probably want to tune the network quite a bit when it comes to multi-gigabits network. There is tons of configurable params with sysctl (like buffer sizes) and ethtool (lro, integrity check) that will impact network speed. I.e. I have a 10Gbps link between my firewall and my desktop, with or without tuning lead to very big difference, like without tuning, it could be stuck at ~4Gbps, with tuning, I was able to get up to nearly full 10Gbps.
Obviously it could be case by case, and could very well be USB chip throttling. However, without tuning the system network params, I think there is also a good chance that the network speed was limited by the OS. Most OS are optimised for 1Gbps and expect multi-gigabit network owner to have the knowledge to perform tuning themselves.
thanks for watching mate and your comments! :) I have 3 networks in my house! 1 x gig and 2 x multigig (2.5 and 10) Although the 10 one is peer to peer with my NAS as I don't need it for anything else. But mate, none of them gave me half the hassle to configure as this device! Big pain in the but! I think it's mostly due to drivers (maybe the device itself)
@@DigitalMirrorComputing I think some OS could be more optimised for 10G, and some requires more configuring.
I am running a OPNsense with CX342, and my desktop running Arch Linux (and Windows) with HP 560SFP+ (Intel 82599ES chip). I found that I run 10G smoothly with Windows, minimal configuration, likely only changing MTU. But require more tuning on Linux.
For the 10G one you have, if you are running Windows to a NAS directly running 10G, probably will be quite easy, as I think both the NAS OS or off the shelf NAS will likely optimise more for 10G as well.
@@DigitalMirrorComputing i have experience with 2.5, 10, 25 and 40gbit. But not with 5. But since 2.5, 5 and 10gbit work exactly the same just with divided signal rate i think the same rules apply. So not reaching full speed gets more of a problem when speed inreases, things like actual buffer space in the hardware and offloading capabilities of the used network chipset on the card become more important. Also the way you connnect the card becomes more important. Although USB is fast enough there is more CPU overhead so on low end devices the CPU might become a bottleneck if you combine it with a USB connected card. Also did you test with just normal frame size of just 1500 because that is the standard every piece of hardware supports in the correct way. Jumbo frames upto 9000 and Super jumbo frames should only be used if you are 100% sure your complete route supports it. If not sure just don’t use it as it can cause all kind of problems in for example vpn software. And you can reach about 4.6 to 4.7gbit with a mtu of 1500 anyway so why bother tweaking with jumbo frames. A mtu of 1500 makes you loose about 3.6% of the theoretical speed with jumbo frames (mtu 9000) you reduce that to 0.6% i refuse to even calculate it for Super Jumbo Frames as like 99% of consumer hardware does not support it correctly anyway. Even in enterprise gear it is not that common.
TrueNas is know for not supporting 5gb and alot of 2.5gb addin cards. So I dropped a 10gb sfp+ dual port nic in my home lab as they are fairly affordable and enuse I has the highest throughput it can for my 2.5gb home network. It might seem like a bit of overkill.....and it is. But it also works beautifully.
didi you put the usb-c stick in at least a 10 gigabit socket. the overhead from the usb protocol might be slowing your transfer speeds down if you are using a 5 gigabit socket
I tried several mate! 10Gb and 20Gb USB-C. Same results! also used an adapter on 2 different computer USB-A 3.0 (with limit of 5Gb) same results unfortunately!
I have used 1 Gbps USB adapters for years on my laptop which doesn't have Ethernet. Works perfectly fine.
However each and every 2.5 Gbps USB adapter I have tried ranged from "occasionally unstable" to "just unusable", even when running on 1 Gbps link.
Those 2.5 Gbps USB ones I tried which were "occasionally unstable", could run stable for hours and sometimes even days but suddenly refused to do anything unless you disabled/enabled the ethernet device or replugged it.
It claimed to be connected, but no traffic possible.
This was about 3 - 4 years ago, so I was hoping they may have fixed those issues as they also were happening with those onboard Realtek 2.5 Gbps adapters.
I clicked your video in the hope to see a positive experience from you with these. For example on a laptop and also for longer periods (e.g. on desktop/server PC)
But given your experience with them, I guess these have still the same kinds of issues as those 2.5 Gbps ones years ago.
My feelings exactly mate! I saw quite a few people talking about these dongles online and giving them stellar reviews but unfortunately that was not my experience! Maybe it was just me, but I feel that the drivers and kernel support just isn't there IMO at least! Maybe you can get better results if you fiddle more with it, but again, IMO that is not a good user experience and hence we can't categorize it as a "GREAT" product. MAx I could get out of it was around 3Gb. And like you point out, on the laptop it was a struggle as well at first!
lovely content !
CAT 7 cable is not what you used because it is a proprietary format and these devices have a normal RJ45 connectors. You probably have been sold a shieled cat 6A that the seller advertised as CAT 7. It could also be a CAT 8 cable but from the seller perspective it would be stupid to under advertised what you sell. Also CAT 8 cables are a pain to use because they are not flexible at all. Also CAT 6 (up to 55m) and CAT 6A (up to 100m) are capable of 10Gbit.
...and yet, they work great with all my other network cards. ACtually you are right, they are CAT 8 !! Sorry about it!
thank you.
Thank you for sharing👍👍👍Mai I know is the USB port on your pc USB3.0 5G?，will 10G USB3.2 port do better?
On my win pc (personal computer) I used a USB-C 10Gb, on the pi I used USB-A which should give you 5Gb. but in neither of those I managed to go over 3Gb! Hope that helps!
@@DigitalMirrorComputing It's possible that the USB-C cable is soldered halfway and to get full speed you need to use the USB 3.2 Gen 2 standard to avoid a bottleneck in the USB connection, although this is unlikely, it's worth checking.
Thanks for the video.
@@Иван_Абрамов could als be that the USB does add a bit more in latency which limits the transfer rate.
If so, then multiple parallel streams may increase total used bandwidth.
Can be tested with iperf for example
To get 5gbe over usb you need to plug into a 10gb USB port, if you plug it into a 5gb USB port the most you’ll get is about 3.5gb, there’s always a overhead.
For the iperf test try multiple streams.
8:12
Its not a "hit or miss" lol. Its skill issue or in this case no dedicated equipment to route (router).
By doing direct connection, you're essentially telling both ends to route which would be fine if it wasn't USB. Getting a multi-gig router or diy pfsense would solve your problems.
well, it sounds like you are a lot smarter than me, so I would say go ahead and get those devices! :) However I must add (for the people reading this) that this comment lacks in (not just courtesy) but accuracy. In a peer-to-peer connection (aka direct link) no router is needed! The two devices can communicate directly with eachother provided (and they are) on the same network! I also don't understand the point around USB... When two devices are connected directly using network adapters (even if they are USB), each device’s network stack handles the direct communication. Routing typically refers to the process of directing traffic between different networks (like from your local network to the internet), and a direct connection does not involve that kind of routing. There are no routing tables involved in a peer-to-peer setup on a single subnet. Routers route traffic! Hope that helps!