[SHARE] IPv6 gigabit solution for now

UPDATED: Not Working.

It may vary by your devices
I got a hAP ac2 that has switch chip (required maybe?)
This is workaround for my IPv6 only network, /64 block from ISP

The concept is using a physical ethernet port to assign an IPv6 prefix

  • ROS 6 only, ROS 7 is even slower for me
  • no default config, no firewall
  • eth1 WAN pppoe to ISP
  • eth2 assign an IPv6 prefix (NOT a bridge interface)
  • eth3-5 as a members of bridge (HW offload by default)
  • connect ethernet cable eth2 to eth3

With this setup i got around 920Mbps for IPv6 by nPerf (single connection)
Only have 2 free ports eth4-5, you can add wlan1-2 to a bridge if you want

I tried assign an IPv6 to a bridge got 800Mbps for few hours then drop to 600Mbps
Comments are welcome

That look promising. I’ve tried something similar on L009 but saw no improvement.
Can you please share your config?

I suspect that is related to the architecture.. Take a look at the block-diagram and how the ports are shared with each other.

I suspect if you use port5 instead of port4 for example, the speeds will drop again.

I gave up and put back in it’s box for months, can’t remember where it is.
It’s not worth the upgrade for gigabit IPv6.

You’re right, speeds drop again and again even unchange any ports.

hAP ac2 and L009 are low-cost low-end products.
In the MikroTik world you cannot expect gigabit performance from those, because all the routing is done in software.
There are other manufacturers who can offer routers at about the same pricepoint that can do it, but they use a different software architecture. Usually they will be much more limited in capabilities and configurability.
When you do not require the extra features and are obsessed with performance, for sure go for those other manufacturers.
When you want to use MikroTik, you have to buy more powerful models to get that performance.

So essentially you are saying that routing IPv6 to a non-bridged interface is a lot faster than routing to a bridge? That’s an interesting observation, sounds like a bug in ROS.

I thought that routing IPv6 to a bridge is a pure CPU process, while non-bridged interface may not because of the test results.
The mistake is my knowledge about routing, all done by CPU.

Thanks pe1chl.

It’s about IPv6 CPU bottleneck that not worth in my opinion. (1 core 100%, 3 cores idle)
While IPv4 (no fasttrack) is running great for all cores and i got gigabit speed with basic or no firewall.

Not all tasks can be distributed over cores.
Especially in “benchmark” scenarios, but also in single-user home usage, you will often see that one core is limiting the performance.
This is not reflected in the test results in the product leaflet, because they are always tests between all ports simultaneously using many different parallel flows.
A single TCP session will not achieve that kind of performance.
So what you need is not a router with more cores, but faster cores.

Thank you for your kindly explain.

My country has generally offer an 1Gbps for home-use at lowest cost.
That mean in my case, I have to buy an enterprise-tier MikroTik for home-use that too huge.

Like you said, there are another home-tier router brands could handle this.

Yes, that is the price you pay for having a 20 year old design that has remained compatible all the time and extended in all possible directions.
Those other manufacturers usually have new models with different features and different firmware optimized for the hardware, and can concentrate on only IP NAT and IPv6 routing.
After 3 years on the market they declare their product obsolete, bring no more new firmware, and move on to the next.
MikroTik is completely different, but that comes at a price.