As market is slowly moving towards 802.11ad (60GHz) for indoor use, 4x4 802.11n/ac implementations and the availability of FTTH becomes broader, dual-core CPUs clocked up to 1.4GHz are also increasingly used in Home and SOHO routers. I just got my hands on a hAP AC and I can’t help feeling that Mikrotik is kinda falling behind in the hardware department, at least CPU-wise.
C’m on guys, single-core 720MHz MIPS overclocked to the max? How much REAL TCP traffic can this chip handle before maxing out, with a few packet mangle rules and filters into play? Ok, I realize that the ARM version of RouterOS is far from ready for serious production use, but you already have a dual-core PPC platform out for a couple of years now, why haven’t you invested more in it?
I absolutely love your work on RouterOS; you rock, you are kings, there is absolutely no contest there, even with the occasional glitches. But the awesome power of your software begs for respectively high power in the hardware (CPU) department. I kinda feel you have neglected that. Maybe you are being held back by specific alliances. If this is the case, you may need to reconsider.
IMHO, you’re failing in the nets of typical SOHO marketing selling tactics.
Have you actually tested the Hap AC instead of comparing specs?
The proof is in the puddle: to provide an example based on the same device you mentioned, Hap AC, four of them were deployed on last Ljubljana MUM conference hall, and each averaged 150 registered stations.
Now try the same with one of those “super-duper” SOHO market routers that will be phased out in three months. Not to mention reliability and stability, firmware continued support, features, etc.
Have in mind MIPS is probably the more optimized hardware platform RouterOS runs on.
ARM platform was probably picked out to be able to compete in the CPU department, Routerboards are aimed at the “Disruptive Technologies” segment, and ARM is clearly the platform with greater mass production (i.e. lower prices) and better projection forecast; Do you really think PPC has any future from this point of view?
Let’s not forget that powerpc is alive and well in other domains, like automotive, where arm and mips have not penetrated yet. So that platform is far from being dead, and more and more powerful cores are available. So I would not dismisst powerpc as a possible midrange platform.
On the other hand, multicore intel atoms with dedicated networking and routing support are also out there, and others are on the drawing board…
I would say, this segment certainly will become more interestig in the near future.
I agree, not saying than PPC is dead, but prices and lineup variety in its offer (also continuous improvements in performance) are driven by actual and prospective sales; and the market ARM is ruling (mobile, etc) isn’t comparable with the rest, let alone automotive which I’d say is on the opposite side: used to incur in costs for components which would be plainly outrageous for the general computing industry; the final price and profit margin are night and day.
One of the starting criteria for Mikrotik when planning for a new product is price; some PPC models price range/features etc will be suitable for mid-line models, and as you both point out is an optimized and stable ROS platform, but I think it will depend on PPC market offer vs ARM and Tilera alternatives.
Supporting two new platforms (Tilera and ARM) implies a huge effort for an embedded device hardware/software manufacturer in this price range market segment, so no doubt Mikrotik has a vested interest on PPC suitability for its lineup; if it’s not happening isn’t due to some sort of conspiracy but lack of suitability and projection.
When I look at the hAP AC, everything about it tells me that its primary intended function is as a simple, but robust access point. As a bridging access point, it has what it needs to forward full-speed AC wireless to/from the wired network. I don’t think it was built to be an all-in-one box that is a FTTH router + access point + fancy firewall + + + + +
It’s the community that looks at the feature set of ROS and thinks that this must mean that this box should be able to work with all of the features turned on. I see the SFP port and 5-port switched ethernet group as “useful in some situations” - not as an open invitation to load this sucker up with as much as possible.
Residential Lines bandwidth is no longer 2 figure numbers, Wireless bandwidth is also increased, rendering the “one router/AP does it all” approach less effective than ever.
Even taking that argument into account, the hAP AC becomes bottlenecked by CPU in AP only mode. The SOHO hardware is definitely lagging behind the competition, the fact that the software is so stable is the only thing that keeps me coming back .
Has MikroTik shared the configuration and RouterOS version they were using on those Hap ACs? I am interested in using the same model for an upcoming conference but am a bit nervous about using such a new device for such an important crowd.
I used the same setup in Bangladesh MUM with 650 people. We had one hAP ac, simply enabled Guest Mode in the Quickset. Basically default config. Worked very well. We had a flawless live video stream though the same device.
Yes, of course. Surely not all people were connected at once. The hAP ac is dual band, so some 70% were using 2GHz, others used the 5GHz. Around 100 were connected at the same time.
The CPU becomes maxed out before the 802.11ac interface reaches maximum throughput. There’s several reports in the hAP AC announcement thread. Hopefully the upcoming newer wireless drivers will improve efficiency.
I’m also hoping that there will be driver and efficiency improvements, especially since upload and download seems to generate a different amount of CPU load.
But as it currently stands, these limits only occur around 500 to 550Mbit of AC wireless throughput, something I’ve only been able to achieve in “lab” conditions. So I do not believe it’s something to worry about. Actual radio connections will almost never be able to reach that limit anyway.
Also, while testing NAT and AC wireless throughput it turns out that even while doing NAT the hAP AC was still capable of doing the 550Mbit AC traffic.
In CAPsMAN mode a bit more CPU would have been nice, using CAPsMAN it’s limited to ~300 to 350Mbit because of CPU.
i guess wAP ac will be the robust AP, and hAP ac will remain the universal workhorse for many.
it performs really nice with 3 figure bandwidth as well.
on the other hand, the up-to-1gbps access speed must be somehow supported by the ISP’s backbone as well.
we do some serious stuff here (with other vendors boxes though) the external links carry 600+gbps peak traffic, but the averaged speed per active customer is still below 1Mbps.
ok, there are bursts, and the device must support it. soon my access will be upgraded to 24ch DOCSIS3.1.
but i think my trusty old RB2011 can handle it. but i’ll have a hAP ac around to see some comparison.
if i could ask something for the so-ho segment, it would be the CPU of hAP ac in a CRS. or a dual core arm in a CRS.
then throw in som wAP acs, and call it a day.
Tests include different packet sizes done with several rules…
So far, even on a device like RB951G it actually exceeds those Xena tests in combination with fasttrack, this on a 60 rules firewall ruleset.
That being said, ROS still needs optimizations to reach CCRs full potential, which ROS v7 will hopefully bring.
Can you provide specific routerboards models / rsc / tests / situations you have done to be able to emit such a taxative conclusion? so far published Xena tests have been very useful for me to size in advance…
And speaking of the CCR1072, 4Gbps you mentioned is even more than the worst-case scenario published figure (64byte packets w/ 25 ip filter rules between 2-3Gbps), those are synthetic tests that need extrapolation for your usage scenario and traffic patterns, just don’t look at the highest number you can locate and hope for the best…
Well I also have problem with maxing out CPU in RB2011 below gigabit - around 68 MB/s, something like that - during NAS access and it’s quite meh but on the other hand… I don’t think it’s that bad for 120$ swiss-knife router+wifi+10/100 switch+everything with touch screen, sfp usb, super-duper OS… so on so on… I’ll probably consider upgrading to some entry level CCR1009. On the other hand I think it’s quite weird that I need to get such tough router for home usage just to reach gigabit to NAS because 3011 is not available in desktop case. Yeah i know i could use switch chip but nah I’m using NAT and firewall between various subnets at home just in case home server would get compromised, so in some places i do have gigabit but not everywhere. So i think CPUs in RB could be a liiiiiitle bit better. Still in love with 2011 as it was my first MT
.
