What do you recommend to purchase a stable RB951G or a new hAP ac2?
after reading this topic viewtopic.php?f=3&t=133001 I doubted the choice.
There is a choice between a stable RB951G and a new, but untested router.
Reread a lot of documentation and made conclusions for myself that for the money in 2018 of course it is better to give preference to hAP ac2, but due to the lack of experience there are questions about the stability of the real range of the wi-fi signal and overheating.

Help with choice, guys

I'm using two RB951G-2HnD units (one as AP and one as router) for a couple of years and IMHO this is a brilliant device if capacity suits needs (somehow lower routing capacity and 2.4GHz-only WiFi).

I just ordered one hAP ac² device and I intend to use it as router. Not that my RB951G units don't give me performance I need but I need a spare RB unit anyway. And I need one to play with (let's call it a lab unit).
To deal with high component temperatures of hAP ac² units I'll drill some vent holes into the case right after "burn-in" period ... to be sure I won't have to RMA it with customized case (I'm sure distributor would rise some eyebrows if I'd RMA a drilled unit). The delay in "hole drilling" project should give me some stats about internal temperatures (I intend to test both lying and standing positions) and I'll have possibility to compare temperatures before and after ... hopefully environment temperatures will not change much during that period (I've got a telecom closet in the cellar).

I'm interested how much different the signal of Wi Fi in comparison with the 951
How much worse is the signal in real conditions, is it critical for the house?

So today I received my new RBD52G (A.K.A. hAP ac²). I did a quick test of WiFi performance in my particular layout (family house, walls made of bricks and concrete, two floors, concrete ceiling between floors). I configured RBD52G in bridge mode, so no routing duties this time. I placed it next to RB951G, which is also configured in bridge mode. All WiFi setup was identical apart from SSID (I wanted to have control over which AP I'm connected to) and 5GHz WiFi on RBD52G. ROS version 6.42.7 on both.
Personally I'm not after last bit of WiFi speed (I'm radio engineer, working for MNO for living, so I believe in hardware - copper and silicon ), so I was focusing on coverage with decent (say 10 Mbps) performance rather than trying to get highest possible speed.

As RB951G doesn't support 5GHz band (and 5GHz has lower coverage after all), I will focus on 2.4GHz performance ...

Specs say that at low speeds (6Mbps or MCS0) RB951G has 3 dB higher transmit power than RBD52G while at higher speed (54Mbps or MCS7) the difference is almost 0. On the other hand, Rx sensitivity is more or less the same at low speeds while it is slightly worse for RBD52G at higher speeds. Which means that RB951G should in theory perform slightly better in DL at low speeds (low signal) and slightly better in UL at high speeds (good signal). As I was focusing on low signal conditions I can't comment on UL performance in good signal.
Now to the observations. I was using my Samsung Galaxy A5 phone with WiFi Analyzer installed. Signal levels showed consistent difference of approx 3dB in favour of RB951G, so specs don't lie. The difference was persistent over whole observed Rx signal strength range (from -55 dB quite near both APs to around -77 dB at the coverage edge). The measured throughput was consistent with this difference. I was using speedtest against internet server, so my DSL line speed (30/5 Mbps) was clearly the bottleneck at decent signal levels. As Rx levels dropped below -70dB, the WiFi became the bottleneck and there observed difference in signal levels showed as difference in acheivable speeds as well. My worst measurements were:
RB951G - signal level -75dB, speedtest result 10/5 Mbps
RBD52G - signal level -77dB, speedtest result 7.5/3.5 Mbps
(both tests conducted at the same spot). Both (difference in signal level and difference in speed) are consistent: difference of 3dB in signal level translates to double (or half) electrical field strength (measured in V/m), 2 dB difference means factor of 1.6 (or 0.6) and 1 dB means factor 1.25 (or 0.8 ) ... the observed speed differ by the same factor.
I can not explain the difference in UL though. Specs say that Rx sensitivity at low signal level should be the same for both APs, so measured UL speed should be the same, but was lower with RBD52G anyway.
Of course I tested the 5GHz AP as well. Specs say that RBD52G has Tx power at low speeds (6Mbps, MCS0) is almost the same as at 2.4GHz. Which I can confirm: close to AP signal levels of both 2.4GHz and 5GHz SSIDs were almost the same (give or take 1 dB). However, wall penetration is much worse at 5GHz than at 2.4GHz, so signal level droped much more when I left the room. And observed speeds dropped with it.

A short note on temperatures: RB951G is not a cool beast either. After a couple of hours, both closed and a small closet and almost idle, both were quite warm to touch. My feeling is that RBD52G is indeed warmer by a few °C (but not much). Case of RB951G has quite a few ventilation holes and my feeling is that after I'll make a few holes in RBD52G case temperatures of both should be almost the same. I imagine that if router's CPU was loaded, the temperature difference would be higher as well (specs say that max power consumption of RBD52G is twice as much as RB951G).

My verdict: if one needs good WiFi coverage, but doesn't need the highest possible speeds, the good old RB951G is still a better choice (and lower price adds to the advantage). However if one needs high speeds, 5GHz ac is a must and RBD52G is a clear winner in this duel. Another use for dual-band APs is small mesh of APs where one band (preferably 5GHz due to higher throughput) is used for "backhaul" ... with single band one simply can not create a decently performing mesh.
As I'm after good coverage at low end speeds, I'll keep using RB951G as AP while RBD52G will move into equipment rack to take over routing duties. I guess temperature will be, despites of CPU being loaded more, still acceptable as WiFi will be disabled.

Meanwhile I did some tests of WiFi performance. RB951G has only 2.4 GHz (802.11n) while RBD52G features both 2.4 GHz (802.11n) and 5 GHz (802.11ac). I tested using iperf pushing UDP streams. I tested using 2 parallel streams as well as 12 parallel streams. Number of streams did not change maximum throughput achieved in any case (except on RB951G), but it did affect CPU load on AP. The wireless client (windows lpatop) was around 3 metres away from APs so signal strength was quite good.

Both APs were configured similarly: 40 MHz on 2.4 GHz using non-overlapping channels (1+5 on one AP and 9+13 on other AP; interference from other APs in area is low). RBD52G was configured to use 80 MHz.

• good old RB951G performed quite well when I used only a few parallel streams. While WiFi connection speed was indicated at 170Mbps, I could push around 105 Mbps in UL and only 60 Mbps in UL. CPU load on RB was around 25% in UL and more than 50% in DL.
  If I pushed 12 parallel streams, UL survived well but DL choked and cumulative DL throughput dropped to less than 20 Mbps. CPU load increased to 100% and clearly CPU is limiting factor when using WiFi on RB951G
• on 2.4 GHz band RBD52G performed slightly worse than RB951G. I could push around 90 Mbps in UL and around 60 Mbps in UL. I attribute worse UL performance to a 5dB lower receive sensitivity of RBD52G at high speed data transfers (MCS7). There's a big difference: RBD52G does not choke on large number of parallel streams, it performs similarly regardless the number of UDP streams. CPU load was around 5% (of a single core) during UL test and around 15-20% during DL test and did not change if number of parallel UDP streams changed.
• on 5GHz band, RBD52G achieved almost exactly double throughput than on 2.4 GHz band in both directions. This is somehow expected as it used double width of frequency channel. I was caught by surprise by observation that CPU load, while shifting double amount of bits, remained the same as during 2.4 GHz tests.
• I tested also throughput, achievable using TCP streams. No surprises here, max achievable throughput was same as with UDP tests, only throughput jitter was somehow more expressed.

Perhaps also interesting was profiling to see which processes consume most CPU cycles on AP while transferring at full speed. As the most CPU-demanding was DL on RB951G I focused on profiling during DL tests. On RB951G cpu cycles are almost evenly distributed between wireless, networking and queuing while on RBD52G majority CPU cycles are distributed between networking and queuing while wireless consumes only a fraction of CPU cycles.

It remains a mistery to me why DL is so much worse on these two devices than UL (DL reaches only 60% of UL). I couldn't identify any bottleneck on RBD52G while RB951G obviously doesn't like many concurrent data streams.

I haven't tested using both WiFi bands (2.4 and 5GHz) concurrently. Considering CPU load during single-band tests I expect, that RBD52G is capable of maxing out on both bands simultaneously.

My verdict: even after speed tests done I still think that good old RB951G can be a god choice if not many devices will consume data in DL ... which is common at home, specially if at least some "bandwidth-hungry" devices are wired. Small office or wireless-only home could be just too demanding for RB951G. In those, more WiFi demanding environments RBD52G may perform quite better. When using it in wireless-only home, one could distribute load between both frequency bands by creating separate SSIDs on both bands and by configuring individual devices to use only one of SSIDs.

Another plus for new RBD52G is switching performace between ethernet ports when one configures VLANs in the new bridge-centric fashion. If configured this way, HW offload is not possible and all traffic between two ethernet ports hit CPU. During 1Gbps test between two ethernet ports, CPU load raised to around 20% (of a single core), but wasn't a bottleneck. So RBD52G in such circumstances does sweat. I can't immagine to configure good old RB951G in same manner, considering its single lower speed core I expect that wire-speed SW switching is not possible on RB951G. Surely one can configure VLANs the old way (using switch chip functionality) on both devices and if done so, RB951G is again back in the game. It's just that configuring VLANs in the old way is not sexy in ROS after 6.41.

Kind of proud for my wAP ac, as it outperforms the newer and much more powerful hAP ac2 even when utilising 2 of 3 chains.
I got stable 200/230 TX/RX with 10 UDP streams, 190/230 with single UDP and 160/200 with TCP.
When testing from iphone I've even seen something like 260-270RX on peaks, but it was less stable, so the average turned out lower.
I wonder how it will perform with three-chained device.
However, no miracles here - the CPU maxes out with multiple UDP both TX/RX, and single UDP TX, and stays around 80-85% for everything else.
On ~150M it feels itself much better with cpu around 30-40%

So I did the last test: I reconfigured RB951G to use new-style bridge VLANs and did throughput test between two ether ports, one being trunk (no tagging/untagging done) and one access (tagging on ingress and untagging on egress). Technically it is switching packets between ports and tagging/untagging on one port. While this device could do 1Gbps wire speed transfer when switch chip did VLAN tasks (tagging/untagging and filtering) and obviously without any effect on CPU load, using bridge VLANs changed things dramatically.

In uplink (PC sending 12 parallel UDP streams), throughput reached slightly more than 910 Mbps while consuming 99% of CPU cycles (almost 40-50% used by networking, the rest 50-60% equaly divided between bridging and ethernet). In downlink, linux server was able to push full 1 Gbps data and RB951G again run at 100% CPU usage (consumption split equal to uplink test).

Verdict: while RB951G was able to switch wire-speed full 1Gbps, it was maxing up CPU, which means it would not be capable of switching full speed between two other ether ports at the same time. Let alone perform some routing duties. RBD52G (hAP ac²), on the other hand, is quite capable of doing wire speed switching with VLAN functionality when bridge VLANs are configured.

So it seems that at least for older devices it is still good idea to use /interface ethernet switch section to configure VLAN functionality if switch's VLAN functionality are sufficient.
At least until Mikrotik implements HW offload of VLAN functions.

RB951G could not quite keep up with my gigabit internet connection with just a couple dozen of Firewall rules and a dozen dst-NAT's. What I could achieve tops around 700 mbit up and 800-820 DL. Direct connection to ISP tops around 940/970 mbit.
Is hAP AC2 stable enough to handle this kind of speeds without being a bottleneck as my old RB951G-2hnd? Could it route 970mbit with simple setup and 20-30 simple firewall rules?
I'm not too interested in wireless speeds (couple of 2.4g phones with light usage) given wireless is just stable and usable for 6-7 MB/s on 2.4G.

I read many articles and topics here with many complaints, heating etc.

My RBD52G is running stable ever since I installed ROS 6.44 and configured VLANs on bridge. I don't know which change did it, but I suspect it's due to not using any non-trivial functionality of its switch chip (seems it has a few bugs).
It is quite warm ... I've got a metallic USB stick plugged in and it is hard to hold it bare-handed, I estimate its temperature to be around 65°C.
Regarding routing/firewalling capacity: with the misery of 30/5 Mbps PPPoE I'm nowhere near your 1Gbps ... but I've never observed CPU load higher than 1%, so I guess it should do full Gbps easily.

Thanks.
Ok, I'm already convinced to buy it after also asking advice in another forum. The guy said he was using some tunnels and VPNs.. on a 100mbit connection and his device used just about 2-4% CPU.
Although Normis said somewhere that the high temps are not to worry about, I'm also almost convinced I'd drill additional holes after few months if the router is still working and I'd just void warranty.
I'll write back with the result and impressions.

But what's wrong with its switch chip? And what do you mean by "non-trivial"?
I'm not using anything special either but if the CPU is powerful enough, could you just turn off hw-offload to bypass the switch chip and use the CPU? Would this bypass any bugs in this chip?
I was using my router without hardware offload for more than a year after I was troubleshooting something, and when transferring in the LAN, of course the CPU was heavily loaded but that wasn't a problem since there are (almost) never two simultaneous transfers between two sets of machines in the lan , moreover I have 10Gbit connection between my desktop and my server so never an issue.

Regarding switch-chip: on paper its features are impressive. However, in my setup I experienced a bug (confirmed by MT support with no ETA for fix), I was experiencing instabillity I already mentioned and another user of this forum reported instability when using non-trivial features of the switch-chip ... wired interfaces become unresponsive and only reboot makes it work again (I've setup watchdog and router would restart 10-20 times a day).

Bug: as mentioned I have PPPoE internet connection (untagged) together with IPTV multicast (tagged). My physical WAN port is cofigured as hybrid port (access port for one VLAN to carry PPPoE and trunk port for tagged IPTV). Switch-chip did not properly untag one of PPPoE initial packets on egress and PPPoE session never established. With exactly same setup, but if connecting ISP device to another device (be it 3rd party managed switch or my good ole' RB951G) PPPoE started just fine.

As you mentioned, RBD52G is a really nice beast (powerful CPU, decent 2Gbps link between CPU and switch chip), so full configuration in software (bridge with vlan-filtering=yes) still allows wirespeed operation. Which is my current config and is running fine.

I suspect most users don't use VLANs so despites using HW offload device performs without problems.

BTW, I drilled an array of holes in the case right above the CPU cooler both on the large side (top when in horizontal position) as well as the smaller side (top when in upright position). While I never did real measurements of temperature, I have a feeling that those do help a bit (more so when RBD52G is in upright position). But then I have a feeling that the metallic USB stick helps even more.

Thanks.
Yes, I won't use VLANs anytime soon (I guess), and definitely not PPPoE (No ISP around ever offered such access and hopefully won't).
What I'm worried about a little, is this 2Gbps link CPU-switch. If WAN port is connected to this chip anyway (as LAN ports), the "internet would have to "flow" twice over this link: from WAN to CPU to handle the traffic, and back from CPU to a LAN port. If there is some overhead here and there, might I expect less-than-ideal wired-speed with this hardware scheme?
I guess tomorrow I'll have the unit delivered to me and test first-hand but I was just speculating.

As to drilling, how did you open it in the first place? Or blindly by inspecting pictures on the internet? I read opening the unit is not straightforward and would damage the casing.
Maybe I can put it under the air conditioner or just use the few months before summer and if still working then I'll drill and void warranty .

My unit arrived.
For now I'm very pleased with its performance (wired). It achieves the same speeds as without a router (~940/960 Mbps Dn/Up) with only about 10% CPU load.

It seems the front "panel" is kind of protruded and along the whole perimeter there is one big ventilation opening that is almost hidden behind the panel projection. Also around the openings for USB and reset button there is a space that also allows for some convection. For now I won't drill anything .

I found a video on youtube where author shows how to open the case ... basically there are two latches holding the front mask (front being the side with ether ports), one on each side, and after unlatching those the front mask comes off ... afterwards you can pull the board out.

The 2Gbps link in RBD52G is twice as much as in other comparable devices. And is enough to allow full Gbps flow from one ether port to another via CPU. I'm just not sure if that link is 2Gbps full duplex or half duplex. I'm sure you'll let us know

How do I test this?

For now I don't find the device too hot, in a room with 25* ambient. I placed it in a tower orientation.

How do I test this?

Perhaps like this: start a big download ... a full DVD ISO (e.g. a Linux distribution) from a speedy source should do. At the same time start your favourite speed test and see if UL speed result is much different than when run without that download running at the same time. If UL speed still reaches almost 1Gbps while maintaining multi-100Mbps in DL, then the CPU-switch connection must be 2Gbps full-duplex. If the UL+DL combined peak at around 1Gbps, then the CPU-switch connection is half-duplex.
But then it is hard to pinpoint where congestion happens ... is it your router that has 2Gbps half-duplex or is it some other box near by (perhaps in your ISP's network) which limits you to 1Gbps half-duplex ...

Note that UL speed test result might show slight slow down which is to be expected ... when doing TCP transmission (e.g. http download), receiver needs to acknowledge received packets. Which might be in form of a TCP ACK packet (packet with only IP and TCP headers, without payload) which is 40 bytes (320 bits) every 2 packets in DL (3000 bytes or 24kbits) assuming no other overhead (PPPoE, VLAN or sone other tunneling). If DL speed is 1Gbps, only ACKs consume almost 13Mbps.

Yeah, but as I expected, my ISP has its "half"-duplex service. If I were to reach ~1Gbps Up and ~1Gbps down, that'd mean 2Gbps internet (which is not true of course), but when I wget-ed a big file with 100MB/s, it plummeted to below 20 and 10 MB/s when I ran speedtest and it got to Upload (which reached 900+ Mbit). I guess at least the device ports are 1000M full duplex and they are capable of 1Gbps up + 1Gbps down.
Otherwise I could have tested any uploud while wget-ing a big file. I have FTP access to a shared hosting allowing 1Gbps transfers.