Not even that… they do not mention the RouterOS version that was used to perform the test. And that can make quite some difference.
Some models even are tested with RouterOS v6 but can no longer run that when purchased today.
I think their testing pages should at least have the RouterOS version added to the tables, and for models that are still being sold there should be separate tables for v6 and v7.
And of course for IPv4 and IPv6 as well, but that would be open for debate. The version number specification isn’t, it is a requirement.
What is the main purpose of a router? routing…
What most people check when comparing different routers? the throughput…
The “Test results” tab on the Mikrotik website show throughput tests. Nice…
But not reachable in IPv6, not even close, not even half, but about the quarter…
My main concern was just, at least, to add a mention that say the tests are done in IPv4 only, which is an important information for a router that is supposed to route IP protocol. Ideally add test throughout for IPv6. But it seems I’m the only one that think it should be the case. So, sorry for the noise, don’t change anything, keep consider that IPv6 is still a POC.
Well, @anav mentioned the “rule of thumb”: when looking at MT official test results, take “Routing → 25 filter rules → 512 byte packet size” number … that one represents real-life performance pretty well (still with a fairly large error margin though). That figure actually represents performance with fasttrack disabled, hence it applies to IPv6 as well. But some of zealots dismissed anav’s remark pretty blatantly.
Yes, MT could be more open as to what benchmark numbers mean. Yes, MT could publish IPv6 results to accompany IPv4 results. Yes, MT could publish ROS v7 results (to accompany v6 results if device can run v6). Yes, MT could take the whole IPv6 story much more seriously than they are now.
And yes, we could discuss al those shortcomings of MT’s test result pages in a civilized manner. But I’m affraid you didn’t come here to discuss things, you came here to bitch about them and you don’t want to accept the explanations we’re offering.
We’re not mindreaders, you know, and neither are people who look at these benchmarks to see if a product meets their requirements.
No, you’re not. I felt let down when I noticed that IPv6 routing is massively slower and haven’t bought another Mikrotik router since. There are other problems with Mikrotik’s IPv6 implementation too, even in the latest ROS7, so I don’t recommend using RouterOS if IPv6 support is important, not just for performance reasons.
Use Mikrotik products for switching and IPv4 routing, like they were intended to be used. Mikrotik’s home country Latvia has very low IPv6 penetration. It is at just 15% now but was practically 0 when the Hex series was new. You don’t order pizza in a Chinese restaurant. Get what they’re good at.
No, but since you came around to discuss things, you could ask about this rule (if you didn’t know where it came from) but you rather decided to be rude towards @anav (“Playing dumb?”)
If I could read his mind and see that he meant that as a rule of thumb, I wouldn’t need to ask for an explanation, but alas, no mindreading. You fanbois are not helping Mikrotik. Every single buyer who gets a router for IPv6 will be extremely disappointed by the IPv6 performance of Mikrotik routers. It is time for Mikrotik to come clean and list realistic benchmarks. I would return the router for a refund if I got it today, and those deceptive benchmark numbers would be the reason.
can you get wirespeed in bridged config? (this eliminates the problem with generation of the traffic)
can you run UDP test instead of TCP (this will eliminate TCP protocol restrictions)
can you run multiple TCP streams? (this will eliminate single CPU core overload)
Bottom line is that single TCP connection is very latency sensitive, so distribute single connection to more that one core will enable TCP slowdown anyway.
It’s not a secret that Mikrotik is not able to handle IPv6 routing in hardware, there is already many topic in the forum about people asking for HW acceleration of IPv6.
For the test I did, it works perfectly in IPv4 with the exact same setup, same router, same computers, etc…
Unfortunately, the IPv6 routing is single-threaded, so when the traffic is reaching the max throughtput, there is 1 core with 100% usage. While in IPv4, with hardware acceleration, 1Gb/s of traffic is using around 25% of 1 core.
Anyway, this is not the objective of this topic, I’m not looking for a technical solution, but asking for a modification on the website to stop showing RB and Hex routers are capable of routing IPv6 at gigabits/s rate while it’s only true in IPv4, or better, show IPv6 real performance.
Agree 100%. Those fancy marketing performance are not helping Mikrotik to stay a clean and trusty company. What do they think? IPv6 is still a world wide POC and will never to mainstream? They hope people will never use IPv6? or won’t notice the ultra low performance in IPv6?
On my “modern” router (RB4011), routing IPv6 at 1Gb/s make 1 core being like 95% used. So it’s fine for my use case. But it also means it can’t do much more, while the router is advertised with a 10Gb/s SFP+ port and a 10Gb/s routing capability…
RB4011 was not released in 2022-2023 (modern). What are you expecting?
But yes the hardware can do a lot more in general, if MikroTik made use hardware-offloaded-XDP/DPDK/VPP. 100G+ networking is possible. But that’s not going to happen with MikroTik. Likely due to legacy codebase of RouterOS, and a complete brand-new code re-write would cost them millions of dollars, which they may not be willing to invest (even Cisco is no different in this regards).
Do we know for a fact that the SOCs in question can even do line-rate hardware IPv6 routing, or is that a wishful presumption based on no better information than “because I waaant it!”
Device, mentioned in first post of this thread, is hEX S and is built around MT7621A. Mediatek’s own propaganda leaflet is mentioning “Network Accelerator” with “2Gbps IPv4/6 routing” … whatever that means.
But then it’s the matter of implementation in ROS … which didn’t even support HW-accelerated VLANs before v7, even less the rest of fancy features. So yes, SoC might support 2Gbps IPv6 routing, but it seems that via some sort of L3HW offload which is not implemented in ROS (and nobody knows if it ever will be). Meaning that in reality, this SoC underperforms in ROS and Mikrotik’s benchmark tests are the only universal way of determining how much it underperforms.
Ages ago Mikrotik changed the routing - it was multithreaded for a single connection, now it isn’t. They did it because the Tilera CCRs were getting latency all over the place, with out of order packets too. This happened because a single TCP stream would use several cores - and sometimes one core got a easier packet than other - hence the “out of order packets” problem.
The routing is still multithreaded - but only for multiple connections. Works like this:
You start (say) an ISO image download, from some HTTP server. Usually it is a single connection for each file. This connection will be routed by a single core, so You will probably be CPU bound - even on a modern router.
Now, imagine that You start downloading 10 ISO images at the same time - from 10 different servers. What we have now are 10 connections - and if your router (edited to change “server” to “router”) has 10 or more cores, each one of them will be routed by a single core.
Your 4011 is advertised as doing 9,7Gbps whith 1512 bytes packages and multiple connections, but no firewall. Yes, multiple connections - they specify it on the table (“All tests are done with Xena Networks specialized test equipment (XenaBay),and done according to RFC2544 (Xena2544)”). RFC2544 dictate multiple connections.
Take a look at the table, and even with fast path and without firewall it doesn’t do 9,7Gbps with 512 byte packets. Now we are talking 7,8Gbps.
Try to route 512 byte with 25 firewall rules and the table says You get only about 2,6Gbps. With, again, multiple connections.
Those SoC used in routers actually allow to accelerate a lot of things, e.g. PPPoE as well.
But to use that, you probably need to use custom software published (with NDA) by the chip manufacturer, and most easily incorporated into a dedicated router program (OS) that also can be licensed. That is what the generic router manufacturers are doing. They sell an almost featureless box that performs at wirespeed.
MikroTik uses Linux and wants to run the same software on many different SoC, so it is more work for them to incorporate all those drivers and optimizations into the software unless they happen to be already in standard Linux. And also, if all that work is spent it makes it more difficult to update the Linux kernel version and it will result in the users complaining about modern Linux features not being in RouterOS.
So it is a difficult decision: do you want top-notch performance from small hardware, or do you want toys like wireguard or wave2/3/4/5 WiFi to be available in RouterOS at the same time ?
From the results I’ve seen posted, it’s impressive what can be achieved with DPDK/VPP. The VyOS project has replaced the use of XDP with VPP for the data plane.
You are trying to make me say something I didn’t say. I expect nothing more from the RB4011. Let repeat myself: “RB4011 […] So it’s fine for my use case.”
I don’t know about line rate, but I know Edgerouter X, that share the same SOC of the Hex S, can offload IPv6 routing. I have one somewhere, if I find it, I test it.
I would like Mikrotik stop claiming their Hex or RB can route many Gb/s of traffic, while in real usage, you have to expect around 10~15% of the announced capacity.
It has been said earlier, but it’s not helping Mikrotik to let them think they can just don’t care about IPv6.
