Thanks! Know if there are any ARM CPUs supported by openwrt that are faster than an i5?
There aren't that I know of.
not sure that hardware even exists at all.
There are some ARM server chips for Amazon/Google/whatever but it's way too big for the job
Seems like it'll be a sweet application for RaspPi 5 - hopefully it has more than x1 PCIe bandwidth for the gigE cards!
There are some notes on 10Gbps networking under generic Linux by the great Jesper Dangaard Brouer in his old blog. These are from 2014, but still 10Gb is still hard when you want to be able to work on small packets....
Can't go beyond 5Gbit because it is connected with a single 10Gbit port to the internal switch. It also has a single 10Gbit physical port.
Even to route that it's most likely using hardware acceleration.
More like Raspi 6 or 7, given how slow is per-core performance increase in commercial ARM processors. Raspi foundation isn't designing their own CPU cores, they just get whatever Broadcomm licenses from ARM. Making a "powerful" ARM server processor is relatively easy, just add a stupid amount of cores and it's fine even if per-core performance isn't that good.
Afaik the only company that is developing ARM cores for this kind of power level ( more powerful than smartphones while also less powerful than massively multicore servers) is Apple, as they are going to replace all their lineup with their new ARM CPUs.
So in a few years you can hopefully do this with a Mac Mini M2 or M3 with a 10Gbit network + a thunderbolt-to-10Gbit adapter. Afaik Linux support for Apple M1 has been merged in kernel 5.13 so you can probably boot OpenWrt in there too in a few years
While I agree on all counts, I wouldn't be so sure about finding 10GBASE-T on Apple gear. Given their history, I wouldn't be surprised at all to see them go wireless-only (one 'ugly' port less for them to bother about, even less two of them).
Surprisingly, 10G has been available as an option on the M1 Mac mini since April. Unsuprisingly it adds a hundred clams to the price. https://9to5mac.com/2021/04/20/apple-silently-updates-m1-mac-mini-with-optional-10-gigabit-ethernet-port/
Says who? And you did hear of the concept of full-duplex before, do you?
Well, yes, that's a different story...
Well, for truly unidirectional traffic you can route ~10Gbps over a single port (e.g. with different VLANs for WAN and LAN), but the moment your load get bidirectional the effectiveness will drop in the worst case of bi-directionally saturating traffic you will only see a gross rate of 5Gbps per direction at best...
The second post in that thread
From the other specifications of this device, this 10Gbit is probably meant as "downlink" port, that allows to move traffic to a switch fast enough to aggregate multiple Gbit connections.
full duplex means "you can send/receive at the same time over the same lines", not "you have separated bandwith for send and for receive".
So you can decide how to split those 10Gbits in send and receive, and for routing the best you can do is an even split, 5Gbit download and 5Gbit upload.
I'm assuming what we want here is routing at 10Gbit up and 10 down is to a 10Gbit switch so that any single device in the local network can access full 10Gbit (just like we do now with 1Gbit). This device is probably capable of routing 10Gbit (with hardware acceleration) between the 10Gbit and all the 1Gbit and 5Gbit ports it has, but this obviously means you pay for 10Gbit internet but each single device is still capped to 1Gbit which is kind of weird, why would you do that. This device is probably going to be stupid cheap for its power, as is tradition with our boy Mikrotik, but it's still not going to be cheap in absolute terms and I'm sure you will still be able to easily buy a random mini itx x86 PC motherboard and a modern i5 or equivalent Ryzen processor and a double 10Gbit card to make a router that can actually handle this level of routing for a similar price if not less.
If you want an example of "you have separated bandwith for send and for receive", that is what PCIe does and it is called double-simplex, aka you have two independent mono-directional communication lines (one for up and one for down) so you actually have two identical but independent bandwith pools. So for example a PCIe x1 v2.0 slot has 500MB/s for up AND 500MB/s for down, it's not 500MB/s total like with full duplex.
Afaik Mac minis and Mac Pros (either the trashcan or the cheese grater) are seen as "professional" use and usually get the 10Gbit ports at least as an option, this has been traditional. The Mac Mini M1 didn't get that at launch because they missed the deadline or Marvell didn't provide them the customized AQtion 10Gbit chipsets with PCIe 4.0 in time, but it was added later (without any official statement, just added to the configurator page). It's Apple, they know everyone would notice within a few days anyway, so why put the effort.
And I say Mac Minis are seen as "professional" because the Minis also have a crappy form of "remote management" (mostly a remote power and reset button) for businness use https://tidbits.com/2021/04/28/m1-based-mac-mini-gets-10-gigabit-ethernet-lights-out-management/
which is similar to the Intel vPro or AMD DASH functionality for workstation PCs.
Laptops and iMacs are meant for the plebs, and they asume plebs don't have 10Gbit or even ethernet at home. Which is not wrong in my opinion.
That's a fair price for what that is. It's a Marvell AQtion AQC133 aka a very new 10Gbit chipset with PCIe 4.0 support, I don't think new 10Gbit cards with AQtion chipsets cost much less than that, and you can be sure new cards with Intel chipsets are NOT cheaper than that either because they are for servers.
Even the used ones on ebay with a single SFP+ port and a modern chipset (like the mellanox CX311A), that are not absolute power hogs are still around 40-50 euros.
joining the discussion with a 10G isp offer (seems legit) https://www.digimobil.es/fibra-optica/pro-digi/10gb and the ONT (modem) user manual (ZTE) https://www.digimobil.es/downloads/Manual_de_Usuario_AX7501.pdf
Yes the next gen Apples will only integrate with your mammary papilla chip, which must be stroked at a genius bar once-per-week to maintain function.
No, you got that wrong. E.g. 10GBASE-T means 20 Gbps full-duplex bandwidth. And there is no choice on splitting its 10 Gbps send and 10 Gbps receive.
I really don't know what you are talking about. Ethernet resp. whatever actual physical connection e.g. by XAU or XFI between MAC and PHY does exactly the same as well. It's full-duplex meaning full 10 Gbps send and 10 Gbps receive.
Each of the 4 Receive and Transmit lanes operates at a rate of 3.125 Gbit/s.
Then, of course, there is some encoding overhead which leaves you with 10 Gbps for receive and another independent 10 Gbps for transmit.
The point about duplex is that a router must take each packet and first receive it, then send it. If a LAN device sends at 10Gbps towards the internet, and the router has a single port, then it will saturate both the send and receive of the router.
If LAN devices in aggregate want to both receive info from the internet, and send to the internet, then they can at most have 5Gbps each direction, because the router must receive and then send each packet.
OK, if the CPU truly would need to look at 100% of the packet contents then that holds true, I agree. That is where hardware acceleration comes into play where one usually talks about control traffic and flow (or streams) of data. Only very few packets need to be analysed and the rest can be assigned to the same flow which will be hardware switched at wire speed.