Fast connection of clients with turris omnia (former: Bonding with turris omnia)

Concerning switch design - will it be possible to enable link aggregation aka bonding for at least one client?

Kind regards,
ssdnvv

Sorry but LACP is not supported. Otherwise connection between switch chip and CPU is only 1Gbit/s. So max. switching and routing speed will be 1Gbit/s

I’m interested in bonding, too.
Can someone say what switch chip is used?

I read that the switch is connected to the CPU via two gigabit ethernet lines.
If the switch supports bonding, or trunking like it is often known as (not at Cisco), it would be possible to use one trunk between CPU and switch, and another between switch and client.

As per openwrt wiki, the older Turris router has a Qualcomm QCA8337N switch chip.
A datasheet (found with google) says this one should support trunking (probably without LACP).

It is a Marvell switch.

Half-knowledge and advanced hacking ahead: :thinking:
If i remember correctly the next best thing should be teaming. If you take into account that you have the added latency of a switch this could work at least with the dsa driver for the switch and a bit of configuration. One CPU port exclusive for the first LAN port and one CPU port shared between the second port for the team and the other LAN ports. Priority should be on first port and second port should only get overflow.

Alternative: use WAN and one LAN port as team and repurpose one LAN port as WAN. This will share WAN and other LAN devices.

Alternate option: add a 2.5 Gbit/s network card using one of the miniPCIe-Slots. 2.5 Gbits for the SFP is imho not possible because of some design limitation but i am not sure.

One question remains: where should the 200 MBytes per second come frome and go to?

Hello
i would be curiose too for this feature
have a nice day
vinc

Perhaps bonding is a relict of the past - I’ve just read an article about NBase-T-standardisation as 802.3bz.
PCIe-Cards for the PC-/server-side connection are already there (Germany: http://www.tragant.de/mail/presse/21_16/21_16_home.html, USA/Canada: https://www.startech.com/Networking-IO/Adapter-Cards/5-speed-pcie-nic~ST10GSPEXNB), what we need is a miniPCIe-NBase-T card to connect a 1/2.5/5/10GBE-machine to our router…

Then we will have:

  • FTTH-speeds via SFP+ (if available in one’s area)
  • fast WLAN (perhaps a MU-MIMO-card (WAVE2?) will be available sometime in the near future to replace the existing chip and make WLAN even faster)
  • fast connection of at least one client
  • most save router-OS available
    That would be the real Swiss army knife-router for me :slight_smile:

Be sure to also take a look at MCTCP.

@adminX: Speaking about speeds this would be a ordinary scenario (in the near future):

  • Laptop uploading backup to server at 1GBit/s via 1GBE-LAN
  • simultaniously watching FHD(/UHD)-videos stored on the server at about 20(-100MBit/s) via WLAN
  • someone else simultaniously
    accessing the servers webserver at 10-40MBit/s via WAN
    In this case you would need at least your mentioned 2,5GBE-Card. Do you know a onlineshop where I could get hold on one of those cards? 5GBE oder 10GBE would even be better…

Even Marvell only describes how to set the board to 2.5 Gbps. As IEEE 802.3bz is just ratified we may have to wait a bit until modules become available.
http://www.aquantia.com/2016/08/17/aquantia-announces-industrys-first-sfp-line-products-enabling-connectivity-2-5-5-10gbase-t-networking/ says Q4 2016 for production release.

One question will be if this works or not. Technical capability of the SoC and all needed connections still do not ensure it will work. Someone will have to take the risk of early adopters and test if this works or not.

If that module you linked to worked and one would reorganize a LAN-port to act as WAN (as in that case the WAN-port would be disabled) we would unfortunately be forced to use an external fibre-modem. But that would only bother those users living in contries that leave the user option to buy their own modems (eg Germany, USA).
But in my eyes that would be a really small flaw - the possibility of using the SFP±port for a fast connection would be really great :slight_smile:

Just being curious: Would it be possible to use a mini-PCIe to PCIe-adapter to connect a PCIe-card? Because it seems not really likely there will come a mini-PCIe-NBase-T-card…

Should be possible and may even be cheaper than the SFP+ module but also looks sometimes a bit ugly. A version with case may look better. Testing with some per-owned PCIe card may minimize the cost to check if it really works.

I think one just needs to be patient atm. It needs some time for the standard to get adopted and for per-owned devices to appear.
There are mini-PCIe-to-PCIe-adapter-cards available for 1GBE for mini-itx boards so there might some similiar cards get developed atm for NBase-T.
Possible connections atm would be:

  • NBase-T-PCIe-cards for about 180€ or 340 US-$ respectively (needed twice, for Links see post above)
  • Adapter without case for about 25 US-$ (->Link)
  • Adapter with case for about 90€ (->Link) or 100 US-$ respectively (->Link)
    So overall costs would be 385-450€ or 705-780 US-$ respectively.
    Both adapter types reach 5GBit/s less overhead (due to USB 3.0-connection-cable), so they would be fine for 5GBE-Connections.

@adminX: What price do you assume the SFP±modules might cost?

Out of the dark i would say the SFP+ modules will cost twice the price compared to PCIe cards and you still have the risk of having nothing.

Trying to find a PCIe network card capable of 2.5GBps SFP+ might be simpler. Connection could then be tried with any of the already available fibre modules or with a SFP link cable.

In the end we won’t know if any of this works as expected until someone takes the risk of losing some money and having some hard time with configuration and kernel development.

[details=My way and off topic]I may go the fibre route and try with some modules. Seems the simplest way as network cards with Linux drivers should be available and SFP+ fibre modules work also on 4GBps Fibre Channel.

Nothing will happen in the near future as there are other things to do:

  • Wait for my Omnia
  • Port ArchLinuxARM to my Omnia
  • Find some missing (NDA-only) information about the Marvell SoC in Omnia
  • Build my Omnia extensions hardware (GPS,PPS,…)
  • Install and setup FreeSwitch
  • Get home automation running
  • Tons of things not related to Omnia…
    [/details]

There is zero overhead. USB3 cables are used because they are cheap and work.

@adminX: Did you already try your fibre route-way?

Well… it won’t work as the mini-PCIe-connector is only PCIe v.1 x1 and therefore only 1GB.

Another thought:
what about this card: https://www.amazon.de/Informatique-CONTROLEUR-MiniPCIe-2-Ports-Ethernet-Dual/dp/B01NBRA4IX/
Its chip supports 2,5Gbit, so it would make up a standalone-LACP-solution? But searching for RTL811 already gives me the hint that the driver is not opensource and therefore has to be built on our own.

Other solutions would be adding one single 1GB-Ethernet-Port:

Best,
ssdnvv

Has anyone yet managed to connect TO faster than 1 GbE?