How does the ePMP compare to the new Ubiquiti Airmax AC system

Products ePMP
25
@RFWaveRider wrote:
@roanwifi wrote:

Hi,

Him forum is plenty of fanboys. It's a shame and very frustating nobody of the "experts" in him forum have never a problem as simple of a poe failure.

As told, if you have a very simple setup, near customers, very good signals, nothing strange, it works. This scenario is very limited in real world.


Also is very frustating the high rate of device failures: flash memory, power supplies, etc.. affecting both APs or CPEs.

I think probably the "fanboys" don't have the problems other people do because they follow best practices.  I've got probably 25-30 Ubiquiti sectors running, with 30 and 40 subs on each one.   We regularlly push 60megabits/second through the sectors with low latency.

I won't install anyone below a -70, and AMC and AMQ need to be above a 75% on the CPEs, as well as the AP AMC and AMQ need to be above 80%.

Are you saying ePMP allows you to just install a bunch of "terrible" clients and everything works well.

That sounds alot like one of their fanboys "It's not the equipment, it's something you did."

It got old having problems with their equipment, trying to get some support, only to get told you don't know what you are doing, the equipment is perfect.

Strange though, how when you would swap their equipment out with another manufacture's equipment and the problem would go away.

To say problems could be avoided with ubiquiti equipment by following best practices is insane.

Ubiquiti equipment has its place but don't confuse it with high quality carrier class gear.

26

Right now, we are working with ubiquti gear in most of our places. Base stations are Rocket M5. No less than -65dbm clients allowed. Recently we introduced iptv, so we thought: next step should be AC gear, it's so much confortable than gps sync and reuse of the channels.

So we bought more than 40 rocket AC (Lite, PTP and PMTP) and their AC antennas. Also, we bought a lot of Nanobeams  AC for new customers... Total disaster.

AC noise requirements for PTMP scenarios are so ideal, that only can be achieved in laboratory. Do you want AC gear? I sold you all my ubiquitis with discount.

27

If you guys are interested in some techical details on how ePMP GPS sync, framing, MAC/PHY improvements work... check out Sakid's technical breakdown of ePMP.

2 Likes
28

Well then I guess I’ve got a laboratory on my towers which are in metro areas.

Sure. PM me a price and we can talk. I’m not saying Ubiquiti is the be all end all, but I’m not seeing the issues you guys are suggesting.

Yes, it’s true, to get AC speeds you need very high SNR.

Frankly, if I lived in a noisy area I wouldn’t even consider trying to run a WISP.

29

I guess there is a quality problem with UBNT. Not sure it is SW only or HW. We once tested M Gear and got horrible results while many people now say it is working well. Not sure if .ac gear needs some more SW releases to get usable or there are HW limits. We see a AC PTMP is about half the price compared to a ePMP Synced AP the Light AP is even cheaper. The nanobeam .ac with 19db is cheaper than the ePMP SM. So may be you simple get what you pay for ...

We tested .ac PTMP with only one AP at a tower and one SM and got very good results with reasonable but not perfect conditions on a 20 MHZ Channel. They still do no ATPC. The webinterface was responsive and simple to use. The ability to see spectrum without disabling the AP is a valuable feature. 

So I guess UBNT ac will shine on low density installations with good signal conditions once they got their SW production quality. With difficult conditions and higher scale ePMP will be better esp. as GPS allows frequency reuse. And above this is the 450 series. 

If ePMP would go .ac and does a 20db panel client UBNT AC would not be considerable ...

30

we’ve ran both systems. before epmp was around, and we looked at markets that couldn’t support the cost of wimax for the cambium 450, we decided to go with UBNT, the gear was ok in low noise, for the $$$ it was great, but rather than “your doing it wrong” or “bad hardware” understanding a systems limits and its behavior in situations can go a long way.
 
can UBNT perform well? sure can. can it beat EPMP is some situations, sure can. But its seems to only win on the bench. For those links I prefer Ethernet ways.
 
back to the root question posted to open this thread, how does ePMP compare to Airmax AC. major point talked about here: .AC AC can code higher, yes. is this viable for a lot of operators? yes and no. if your rural, with little noise and small distances to cover then this benefits you. if noise is an issue, or you’ve got longer distances to cover, than AC likely has little to no benefit to you. I’ll take a moment to point out the 450 has better SNR requirements and link budgets than AC gear if your after raw performance without the cost of LTE
 
next highly talked about item I’ve seen in this thread: Antenna gain. gain is important, but you’ve skipped over some major, major points to consider. the AIR max sectors have an usually small vertical lobe, which if you understand the antenna specifications completely, you can twist that to your advantage. simply mount and forget a tight antenna and you’ll be scratching your head about signal problems down the road. 2nd if you take a moment and look over the radiation graph and notice the “roll off” ubnt antennas are scored at -6 db on the edges, what does that mean exactly? this means an antenna that is rated at 20 DB forward power and is a 90 degree sector, your antenna gain at the azimuth edge is 14db. so a customer who is situated in-between sectors don’t get 20 db of antenna to work with, they get 14. and if your following system rules, only 30 DB ERIP when your legally allowed 36 in the upper portions of the band. a lower gain, but more evenly radiated antenna will result in 33 DB ERIP being aimed at that subscriber, resulting in 3 DB MORE POWER to that subscriber from the weaker antenna. next you need to look at lobe match. how well does the vertical and horizontal lobes match. some antennas can be as much as 5db off at lobe edges but still claim to be a 20 db antenna. but one of the lobes near an edge is 14 vertical and 11 db horizontal? in short, the antennas effective radiation of energy over a give area as advertised can be far more important than those 2 or 3 DB of extra power you think your getting, when infact, on a sector edge that same antenna can be weaker or worse, miss matched power on your radio chains.
 
next major thing to consider between the radio platforms: the PHY and MAC layers. how do they work exactly. UBNT still relies on polling, how does polling work? polling works by talking to each sub scriber in order. basestaion will TX to CPE, wait the for a ack or Nack and either repeat or move to the next sub, wash rinse and repeat. for a few subscribers, with a small about of re-transmit, this works very well. and will actually show lower latency than the EPMP system. the draw backs, since the AP talks then listens to each CPE 1 at a time, and retires frames before moving on, if you have a bad RSSI or poor modulation with a given CPE, the AP will spend a considerable more amount of time to talk that CPE . this is bad in the sense that it can spike latency and reduce bandwidth to ALL CPES connected to that AP. if the AP keeps having to ask CPES to retransmit frames from AP side noise, a major amount of your air time can be lost to simply askin if the packet was sent successfully. those problems can be prevented from maintaining very good SNR, very high link qualities ect. but one bad onion and all STAs suffer. next issue is the issue of wifi based devices checking to see if the environment is clear before talking, adding more delay and if noisy the device literally sitting quite before it talks.
 
the EPMP PHY and MAC layers work very differently. the AP schedules time slots from ALL cpes to interact with it, that also applies to the retransmit frame request. this causes latency to be higher than the standard wifi mac, you’ll see 15 to 20 ms pings, and 20 to 30 when loaded. but you’ve got a number of advantages to go with the small latency trade off. a scheduled frame divides up air time across all of the CPEs wanting to talk or receive information during the frame. the access point will start the frame by sending out a schedule of what is about to happen. it will deliver the data to CPEs it has to send next, followed by listening for responses from CPEs in the order and time slot it has pre-defined for the CPES. this also brings into the factor of being able to connect a weaker CPE without a detriment impact of performance. in short, since time is the factor rather than throughput you can connect weaker subs and sell them less speed and have that sub and with other systems, connecting that sub will result in a crash across the board with that AP. do I advise doing that? not if you can help it, but its a card you have to play.
 
what does this mean to the operator? this means that a poor SNR sub, or a sub that is plagued with signal problems doesn’t break other subscribers connections. it also allows for tighter QOS control, it also allows for many sessions to be active together with calmer changes in latency, it also isn’t bother with the listen for gap and talk issue that wifi has.
 
the next major thing EPMP does over other wifi based system is dodging the bullet of bursty radio traffic. this mean if the base station is sending all frames in MCS15 but is having to retransmit to many frames, to the point that shifting to MSC 14 without retransmits would be more effect, if will do so. the modulation isn’t solely based on SNR. its based on success rate as well which significantly increase throughput when dealing with bursty noise. best of my knowledge, AC is still based on SNR, rather than also taking into account the amount of frame retransmission rate. so the AC gear has a change to out modulate, but also runs the risk to send far more retransmits running the risk less useful air time and boiling down to less useable bandwidth. since airmax doesn’t take into account the retransmit rates, your taking a guess on that aspect.
 
another things to consider is entire system performance: how many 40 MHZ channels do you have to work with in the upper band? 4. wouldn’t it be great to have a tower every 3 miles and have 4 APs on each tower bringing you 130+ meg of downlink capacity on every tower just a few miles apart? you could actually keep up with the phone and cable company if you wanted to. can you use those same 4 channels that close together without GPS? the answer is simply no. the AC argument, higher coding rates, 30% is your speed increase, but with EPMP you can use the same channel twice, with panels back to back. that’s a 100% increase over N spec. or how about the guys in the sticks with 2.4 GHZ radio? wouldn’t it be nice to have 6 access points on 2.4 on the same tower running 20 MHz channels? granted AC in 2.4 won’t happen but its there with EPMP if you want it.
 
personally the fail rate we experienced from UBNT is insane to my standard. we replace almost double the UBNT gear as we do all other vendors combined and it makes up only 15% of our network. that says a lot to me. thusfar the EPMP seems to be as reliable the PMP100 stuff. solid. to re-cap.
 
if you can lock every sub at -65, you have little noise, and a low sub and AP density, say 15 subs a sector, and 3 sectors a tower ( plenty of guard band and don’t forget your RF armor shields! ), and just a tower or two. you’ll likely see little difference from EPMP over UBNT. however if you want density, 4 or 6 APs a tower, 40 to 60 subs a panel, and towers near each other and plenty of them plus you have noise and can’t lock every sub at -65, well then you’ll see a major difference between cambium and UBNT. don’t forget fail rate and service call rates. most service providers who has run wifi and something else can tell you nightmares about service call rates. feel free to hit me with questions, I dropped a lot in this post.

6 Likes
31
@Chris_Bay wrote:

we've ran both systems. before epmp was around, and we looked at markets that couldn't support the cost of wimax for the cambium 450, we decided to go with UBNT, the gear was ok in low noise, for the $$$ it was great, but rather than "your doing it wrong" or "bad hardware" understanding a systems limits and its behavior in situations can go a long way.

can UBNT perform well? sure can. can it beat EPMP is some situations, sure can.

back to the root question posted to open this thread, how does ePMP compare to Airmax AC. major point talked about here: .AC AC can code higher, yes. is this viable for a lot of operators? yes and no. if your rural, with little noise and small distances to cover then this benefits you. if noise is an issue, or you've got longer distances to cover, than AC likely has little to no benefit to you. I'll take a moment to point out the 450 has better SNR requirements and link budgets than AC gear if your after raw performance without the cost of LTE

next highly talked about item I've seen in this thread: Antenna gain. gain is important, but you've skipped over some major, major points to consider. the AIR max sectors have an usually small vertical lobe, which if you understand the antenna specifications completely, you can twist that to your advantage. simply mount and forget a tight antenna and you'll be scratching your head about signal problems down the road. 2nd if you take a moment and look over the radiation graph and notice the "roll off" ubnt antennas are scored at -6 db on the edges, what does that mean exactly? this means an antenna that is rated at 20 DB forward power and is a 90 degree sector, your antenna gain at the azimuth edge is 14db. so a customer who is situated in-between sectors don't get 20 db of antenna to work with, they get 14. and if your following system rules, only 30 DB ERIP when your legally allowed 36 in the upper portions of the band. a lower gain, but more evenly radiated antenna will result in 33 DB ERIP being aimed at that subscriber, resulting in 3 DB MORE POWER to that subscriber from the weaker antenna. next you need to look at lobe match. how well does the vertical and horizontal lobes match. some antennas can be as much as 5db off at lobe edges but still claim to be a 20 db antenna. but one of the lobes near an edge is 14 vertical and 11 db horizontal? in short, the antennas effective radiation of energy over a give area as advertised can be far more important than those 2 or 3 DB of extra power you think your getting, when infact, on a sector edge that same antenna can be weaker or worse, miss matched power on your radio chains.

next major thing to consider between the radio platforms: the PHY and MAC layers. how do they work exactly. UBNT still relies on polling, how does polling work? polling works by talking to each sub scriber in order. basestaion will TX to CPE, wait the for a ack or Nack and either repeat or move to the next sub, wash rinse and repeat. for a few subscribers, with a small about of re-transmit, this works very well. and will actually show lower latency than the EPMP system. the draw backs, since the AP talks then listens to each CPE 1 at a time, and retires frames before moving on, if you have a bad RSSI or poor modulation with a given CPE, the AP will spend a considerable more amount of time to talk that CPE . this is bad in the sense that it can spike latency and reduce bandwidth to ALL CPES connected to that AP. if the AP keeps having to ask CPES to retransmit frames from AP side noise, a major amount of your air time can be lost to simply askin if the packet was sent successfully. those problems can be prevented from maintaining very good SNR, very high link qualities ect. but one bad onion and all STAs suffer. next issue is the issue of wifi based devices checking to see if the environment is clear before talking, adding more delay and if noisy the device literally sitting quite before it talks.

the EPMP PHY and MAC layers work very differently. the AP schedules time slots from ALL cpes to interact with it, that also applies to the retransmit frame request. this causes latency to be higher than the standard wifi mac, you'll see 15 to 20 ms pings, and 20 to 30 when loaded. but you've got a number of advantages to go with the small latency trade off. a scheduled frame divides up air time across all of the CPEs wanting to talk or receive information during the frame. the access point will start the frame by sending out a schedule of what is about to happen. it will deliver the data to CPEs it has to send next, followed by listening for responses from CPEs in the order and time slot it has pre-defined for the CPES. this also brings into the factor of being able to connect a weaker CPE without a detriment impact of performance. in short, since time is the factor rather than throughput you can connect weaker subs and sell them less speed and have that sub and with other systems, connecting that sub will result in a crash across the board with that AP. do I advise doing that? not if you can help it, but its a card you have to play.

what does this mean to the operator? this means that a poor SNR sub, or a sub that is plagued with signal problems doesn't break other subscribers connections. it also allows for tighter QOS control, it also allows for many sessions to be active together with calmer changes in latency, it also isn't bother with the listen for gap and talk issue that wifi has.

the next major thing EPMP does over other wifi based system is dodging the bullet of bursty radio traffic. this mean if the base station is sending all frames in MCS15 but is having to retransmit to many frames, to the point that shifting to MSC 14 without retransmits would be more effect, if will do so. the modulation isn't solely based on SNR. its based on success rate as well which significantly increase throughput when dealing with bursty noise. best of my knowledge, AC is still based on SNR, rather than also taking into account the amount of frame retransmission rate. so the AC gear has a change to out modulate, but also runs the risk to send far more retransmits running the risk less useful air time and boiling down to less useable bandwidth. since airmax doesn't take into account the retransmit rates, your taking a guess on that aspect.

another things to consider is entire system performance: how many 40 MHZ channels do you have to work with in the upper band? 4. wouldn't it be great to have a tower every 3 miles and have 4 APs on each tower bringing you 130+ meg of downlink capacity on every tower just a few miles apart? you could actually keep up with the phone and cable company if you wanted to. can you use those same 4 channels that close together without GPS? the answer is simply no. the AC argument, higher coding rates, 30% is your speed increase, but with EPMP you can use the same channel twice, with panels back to back. that's a 100% increase over N spec. or how about the guys in the sticks with 2.4 GHZ radio? wouldn't it be nice to have 6 access points on 2.4 on the same tower running 20 MHz channels? granted AC in 2.4 won't happen but its there with EPMP if you want it.

personally the fail rate we experienced from UBNT is insane to my standard. we replace almost double the UBNT gear as we do all other vendors combined and it makes up only 15% of our network. that says a lot to me. thusfar the EPMP seems to be as reliable the PMP100 stuff. solid. to re-cap.

if you can lock every sub at -65, you have little noise, and a low sub and AP density, say 15 subs a sector, and 3 sectors a tower ( plenty of guard band and don't forget your RF armor shields! ), and just a tower or two. you'll likely see little difference from EPMP over UBNT. however if you want density, 4 or 6 APs a tower, 40 to 60 subs a panel, and towers near each other and plenty of them plus you have noise and can't lock every sub at -65, well then you'll see a major difference between cambium and UBNT. don't forget fail rate and service call rates. most service providers who has run wifi and something else can tell you nightmares about service call rates. feel free to hit me with questions, I dropped a lot in this post.

Regarding higher density AP installations UBNT tries to address this with some filters to increase Adjacent Channel Rejection. This is not included with the Light but with the more expensive PTMP AP and comes with a limit to max 40MHz Channels. There are no technical numbers published how this filtering works. A UBNT user did some testing:

http://community.ubnt.com/t5/airMAX-Stories/Rocket-AC-ptmp-Ultimate-airPrism-Adjacent-Channel-Attack/cns-p/1188160

High quality gear like 450 has a higher Adjacent Channel Rejection by design which allows to colocate without guardband.

Do you have deeper information on airmax .ac? This is different/incompatible to the airmax of .11n based gear. I did not seen any publication other than sales stuff. Cambium does a much better job explaining what is happening over the air.

The failure rate is a real concern. We see this with Mikrotik, too. This cheap SXT Stuff is dying more often. So the lifetime has to be calculated into the price. This week we had a 1 year old CCR dying which is a bigger problem as this are core devices.

1 Like
32

UBNT  promotes it's  AC gear  as new highly efficient system, that outperforms all 802.11N  based system ( also it's old airmax 802.11N ) and competitors.

Let's see   how ubnt АС really works.

1) Point To Point scenario.

АС chip has 256QAM5/6 modulation and is able to achieve max 135-140 MBps ( UL+DL) in 20 MHz channel bandwidth, 270-280 Mbps in 40 MHz.  Wi-Fi AC chip requires for 256QAM5/6 modulation  CINR =signal/( interference+ termal noise ) >34dB. Taking into account 3 dB fade margin required  CINR for stable link at 256QAM5/6  should be 37 dB. It means that when interference + noise is -85 dBm ( typical interference at base station locations in 5 GHz), then signal should be > -48 dBm. So max distance, for example, PB AC 500 ( antenna 27 dBi, and Tx 18 dBm@ 256QAM5/6)  in real life noisy environment (about -85 dBm)  is  3km  ( it is able to achieve  max 135-140Mbps in 20 MHz at 256QAM 5/6 ).

UBNT AC shows (at web) measured by AC radiochip it's (interference + noise) level and CINR, but ubnt system  can not adapt it's modulation according measured CINR. And when distance is longer then 3 km or interference is higher then -85 dBm ubnt АС tries to works at 256QAM5/6 at CINR lower then 37 dB. For example, we very often may see that ubnt AC works  at 256QAM 5/6 at indicated at it's web CINR 26-29 dB that leads to huge packet losses  and big TCP/IP throughput degradation.

See screen of TCP duplex test, that made by Bandwidth Test Mikrotik for link 13km at PB AC 500, 40 MHz,  antenna dish 27 dBi, (interference+noise)= -86 dBm, CINR 23-29 dB, but PB works at 256QAM/64QAM

18-180-tcp.jpg
18-180-tcp.jpg1562×798 370 KB

It was got 170M download +18M upload TCP duplex. Test was done by experienced ubnt fanboy , and  it was best result ,that PB AC 50 was able to show at 13 km in such condition. If download to remain 170Mbps and upload to increase up to 20 Mbps, then link latency became over 80-100 ms.  So AC at 256QAM 5/6 shows  in 40 MHz TCP 188 Mbps ( UL+DL). It is  much better then old Airmax N ( for example PB M5 400) in the same link, but it is almost the same as  ePMP Force with 25 dBi dish  is able to show TCP 170+20 Mbps at 64QAM 5/6, 40 MHz  in the same conditions. But ePMP also is able to achieve symmetric dupleх 100+100М, when UBNT AC – not ( when current CINR less required CINR).

So main problems of ubnt AC  are high required CINR for 25QAM5/6 and luck  of ability to manage it's modulation according current CINR.

Second problem is Airmax AC high packet error rate (PER). For example we may see in UBNT AC big difference between UDP test throughput ( that does not take into account packet losses) and TCP test throughput, that degrades when  packet losses occur.  Both problems  are  Airmax protocol issues, and it may be resolved only by  complete Airmax redesign, that is impossible.

2) Point to multipoint.

As we know UBNT Airmax N  is able to achieve in 20 MHz point to point  max 90-100 Mbps. In multipoint scenario with 25-30 Subscribers ( 6-8+ simultaneously active SS)  ubnt sector  throughput capacity is 22-25 Mbps ( UL+DL) , in 4 times less then in p2p. Also we know, that  ubnt  is able to serve with acceptable quality max  30-35 CПЕ per sector ( with  max 25 Mbps sector throughput).

If AC able to achieve 140 Mbps at 256QAM 5/6 in 20 Mhz  in p2p, we may suppose, that in multipont with 30 SS it may give only 35 Mbps. Why AC should give more then 35 Mbps in 20 MHz  multipoint throughput? Why ubnt AC  should serve more then 30 СПЕ sector Airmax N ? It seems, Airmax AC multipoint protocol remains the same as Airmax N, so nothing is changed and all problems are the same.

Let's assume , that  ubnt   Rocket AC  and all CPE  ( PB AC ) 25-30 q-ty  are working in multipoint at 64QAM modulations , -  at the same  modulations as,  for example, ubnt  AP Rocket Ti with CPE - PB M5. Total sector throughput of Rocket AC and Rocket Ti with 25-30 CPE at 64QAM will be the same - about 25 Mbps (UL+DL). Will total sector throughput increase in twice-   till 50 Mbps, if  all CPE AC , that are working with Rocket AC AP increase it's modulation up to 256QAM ? No. Max sector throughput capacity growth  due to  increasing CPE  modulations from  64QAM to  256QAM will be  about 30-40%. So it will be  max 35 Mbps. May be 40 Mbps, no more

Somebody may say, that Rocket AC has ASIC  Airmax accelerator. But throughput  and max quantity CPE per sector  does not depends on CPU power. So ASIC can not significantly improve UBNT AC Airmax performance.

We have never seen yet in multipoint above 30 CPEs per UBNT AC sector and above 30-35 Mbps sector throughput with 25-30 active CPEs.  All that was shown by UBNT fans in their forum, it is some kind of p2p throughput, when 20 CPE are idle, and only one CPE is working and shows point -to point speed.

Also if we take into account how many CPEs are able to work in multipoint at 256QAM5/6 ( because for 256QAM CPE should have 27-30 dBi dish at 2-3 km LOS distance from base station) we can understand  that UBNT AC gear is not able to improve UBNT multipoint service.

In comparison with ePMP multipoint. We know and we have examples, when ePMP 20 MHz sector serves 120 CPEs with throughput sector capacity total 70 Mbps download + 8 Mbps upload ( MRTG daily graph) So conclusion is:

- in point to point UBNT AC is able to compete with ePMP  in throughput  only with high gain dish antenna  at short distance  in low noisy environment. 

 - in multipoint  UBNT AC  is not able to compete with ePMP in total sector throughput and max CPEs per sector.

2 Likes
33

Hi,

just one question:

>>АС chip has 256QAM5/6 modulation and is able to achieve max 135-140 MBps ( UL+DL) in 20 MHz channel

For real customer trafic ~135-140 Mbit/s (eg. 120 download, 20 upload), this translates to ~20.000pps

Have you performed any test on such link, adjusting the packet size or so in order to have this PPS figures?

It's clear the AC is able to pass 135 or 140 or 300 Mbit/s with the biggest packet size (1500+) and with packet aggregation in the PHY, but is able to pass such amount of packets?

Regards

Antonio

34

roanwifi wrote:

For real customer trafic ~135-140 Mbit/s (eg. 120 download, 20 upload), this translates to ~20.000pps

Have you performed any test on such link, adjusting the packet size or so in order to have this PPS figures?

It's clear the AC is able to pass 135 or 140 or 300 Mbit/s with the biggest packet size (1500+) and with packet aggregation in the PHY, but is able to pass such amount of packets?

______________

Average packet size of live TCP/IP traffic in real networks is about 700-800 Bytes. So it means 140 Mbps  live traffic  has about  20K pps.

Ubnt Airmax 80211.N pps depends on packet aggregation   level at MAC  and   with max A-MPDU 64k bytes aggregation  UBNT 11.N Airmax  max pps  is  20-22K. In noisy environment with lower aggregation level UBNT M5 has  usually about  17 K pps .

ePMP in ePTP mode  has 25K pps  and it does not depends on A-MPDU aggregation, because ePMP MAC  has not it.  So ePMP is able to pass in 40 MHz real traffic ( about  800 bytes packet size)  only   150Mbps ( UL+DL).  Max ePMP througput 220 Mbps (UL+DL)  in 40 MHz is able to be  achieved  at packet size 1500 bytes ( for example at FTP traffic).

If packet size is smaller then throughput is  also reduced.  For  example uTorrent may have minimum  150 bytes  UDP packet size , and ePMP may pass  only 30 Mbps of such Torrent traffic.

802.11AC  hardware has more pps then 802.11N .  I do not know  it’s  pps precise value , but  it definitely has  more then >40 K pps . So AC does not have 150 Mbps limitation of real traffic throughput as 802.11N  MIPS hardware has.  Its  may be one and only one  advantage of AC wave1 hardware for outdoor BWA applications in comparison with 802.11n . 

I belive, that  next generation  of ePMP  ( also based on 802.11N radio ) will have higher pps ( >40-50K pps is desireable ) and we can get full 220 Mbps throughtput in 40 MHz at live real traffic with small packets size.

35

>>Ubnt Airmax 80211.N pps depends on packet aggregation   level at MAC  and   with max A-MPDU 64k bytes aggregation  UBNT 11.N Airmax  max pps  is  20-22K. In noisy environment with lower aggregation level UBNT M5 has  usually about  17 K pps .

Are we talking about same ubiquity m5 airmax? rocket m5? BY FAR I have never seen such figures even on the most ideal signal/better noise level. For PtP < 8k-9k. For PtMP < 2-3k. And these figures are very dependant on distance, due the ACK protocol, if distance increases, the airtime silences increases and  performance decrease...

36
@Vyacheslav wrote:

@roanwifi wrote:

For real customer trafic ~135-140 Mbit/s (eg. 120 download, 20 upload), this translates to ~20.000pps

Have you performed any test on such link, adjusting the packet size or so in order to have this PPS figures?

It's clear the AC is able to pass 135 or 140 or 300 Mbit/s with the biggest packet size (1500+) and with packet aggregation in the PHY, but is able to pass such amount of packets?

______________

Average packet size of live TCP/IP traffic in real networks is about 700-800 Bytes. So it means 140 Mbps  live traffic  has about  20K pps.

Ubnt Airmax 80211.N pps depends on packet aggregation   level at MAC  and   with max A-MPDU 64k bytes aggregation  UBNT 11.N Airmax  max pps  is  20-22K. In noisy environment with lower aggregation level UBNT M5 has  usually about  17 K pps .

ePMP in ePTP mode  has 25K pps  and it does not depends on A-MPDU aggregation, because ePMP MAC  has not it.  So ePMP is able to pass in 40 MHz real traffic ( about  800 bytes packet size)  only   150Mbps ( UL+DL).  Max ePMP througput 220 Mbps (UL+DL)  in 40 MHz is able to be  achieved  at packet size 1500 bytes ( for example at FTP traffic).

If packet size is smaller then throughput is  also reduced.  For  example uTorrent may have minimum  150 bytes  UDP packet size , and ePMP may pass  only 30 Mbps of such Torrent traffic.

802.11AC  hardware has more pps then 802.11N .  I do not know  it’s  pps precise value , but  it definitely has  more then >40 K pps . So AC does not have 150 Mbps limitation of real traffic throughput as 802.11N  MIPS hardware has.  Its  may be one and only one  advantage of AC wave1 hardware for outdoor BWA applications in comparison with 802.11n . 

I belive, that  next generation  of ePMP  ( also based on 802.11N radio ) will have higher pps ( >40-50K pps is desireable ) and we can get full 220 Mbps throughtput in 40 MHz at live real traffic with small packets size.

I guess if cambium would want to build a next generation of ePMP with 11n they run into problems getting a faster 11n chipset. Chipvendors went to .ac already. They wont invest in building another 11n chipset. And so would cambium.

Dont try to neglect the newer .ac chipsets are better than 11n chipsets. They have faster processing and they achieve higher modulations at short distance. The fact that UBNT AC do not perform well enough now does not mean that the hardware base is worse compared to older HW. I guess if cambium would use this HW base and add GPS it would perform and scale quite well.

37
@ste wrote:
I guess if cambium would want to build a next generation of ePMP with 11n they run into problems getting a faster 11n chipset. Chipvendors went to .ac already. They wont invest in building another 11n chipset. And so would cambium.

 Dont try to neglect the newer .ac chipsets are better than 11n chipsets. They have faster processing and they achieve higher modulations at short distance. The fact that UBNT AC do not perform well enough now does not mean that the hardware base is worse compared to older HW. I guess if cambium would use this HW base and add GPS it would perform and scale quite well. 

The way I look at the 802.11n vs 802.11ac debate is that we've seen what Cambium can do with an 802.11n chipset... amazing things! Higher performance then the vast majority of competitor's 802.11n based products, great PtMP scaling, WiFi backwards compatibility and migration tools, actual working GPS sync and frequency reuse which is a feature that I don't believe any other manufacturer has been able to successfully accomplish on an 802.11n chipset, and the list goes on... Now imagine what they could do with a mature 802.11ac chipset... yes, it might take them a bit more time for a new 802.11ac ePMP platform to be developed, but I think they want to avoid releasing a half-baked product... they want to build on the success of the 802.11n ePMP platform. I think that's something worth waiting for.

1 Like
38

I guess if cambium would want to build a next generation of ePMP with 11n they run into problems getting a faster 11n chipset. Chipvendors went to .ac already. They wont invest in building another 11n chipset. And so would cambium.

Dont try to neglect the newer .ac chipsets are better than 11n chipsets. They have faster processing and they achieve higher modulations at short distance. The fact that UBNT AC do not perform well enough now does not mean that the hardware base is worse compared to older HW. I guess if cambium would use this HW base and add GPS it would perform and scale quite well.

____________________

ePMP 802.11n chipset is manufactured by Atheros-Qualcomm.   In next generation ePMP (  based on 802.11n  radio) will be used  the same  Qualcomm chipset. But new ePMP hardware ( board with CPU, and also  radio) will be more powerful and efficient.  We believe, that new ePMP platform will  have increased pps, higher MAC TDMA  and  radio performance.  And current  ePMP now  and next generation  802.11n based ePMP  are (and will be)  more efficient ( more max CPEs per sector,  more sector throughput, lower and stable latency and jitter, lower PER. etc. ) then any N and  AC chip based systems from other vendors.

Ubnt AC is based on 802.11 AC wave 1 chipset.   They say that  AC wave 1 chip  is not good enough for outdoor BWA product. It seems new ePMP AC will be based on 802.11 AC wave 2 MU-MIMO chipset ( I suppose , it will be also Qualcomm).  At the beginning ePMP AC will be   Point-To -Point.   In regards to AC multipoint, I would say, that    AC  in multipoint  may be efficient  do not due to using 256QAM ( it requires high  CINR, that it is hard to achieve in real multipoint network), but due to MU-MIMO. It means that users ( CPEs) in multipoint are separated by  timing ( TDMA)  and  also by using different polarizations. It  should increase  efficiency of  multiple access.  May be sometime we will see also Cambium ePMP AC with TDMA MU-MIMO  multiple access for  multipoint networks. And  if  it  happen,  it will be  completely other product ( not only GPS sync will be advantage)  in comparison  with UBNT  AC.

1 Like
39
@Vyacheslav wrote:

I guess if cambium would want to build a next generation of ePMP with 11n they run into problems getting a faster 11n chipset. Chipvendors went to .ac already. They wont invest in building another 11n chipset. And so would cambium.

Dont try to neglect the newer .ac chipsets are better than 11n chipsets. They have faster processing and they achieve higher modulations at short distance. The fact that UBNT AC do not perform well enough now does not mean that the hardware base is worse compared to older HW. I guess if cambium would use this HW base and add GPS it would perform and scale quite well.

____________________

ePMP 802.11n chipset is manufactured by Atheros-Qualcomm.   In next generation ePMP (  based on 802.11n  radio) will be used  the same  Qualcomm chipset. But new ePMP hardware ( board with CPU, and also  radio) will be more powerful and efficient.  We believe, that new ePMP platform will  have increased pps, higher MAC TDMA  and  radio performance.  And current  ePMP now  and next generation  802.11n based ePMP  are (and will be)  more efficient ( more max CPEs per sector,  more sector throughput, lower and stable latency and jitter, lower PER. etc. ) then any N and  AC chip based systems from other vendors.

Ubnt AC is based on 802.11 AC wave 1 chipset.   They say that  AC wave 1 chip  is not good enough for outdoor BWA product. It seems new ePMP AC will be based on 802.11 AC wave 2 MU-MIMO chipset ( I suppose , it will be also Qualcomm).  At the beginning ePMP AC will be   Point-To -Point.   In regards to AC multipoint, I would say, that    AC  in multipoint  may be efficient  do not due to using 256QAM ( it requires high  CINR, that it is hard to achieve in real multipoint network), but due to MU-MIMO. It means that users ( CPEs) in multipoint are separated by  timing ( TDMA)  and  also by using different polarizations. It  should increase  efficiency of  multiple access.  May be sometime we will see also Cambium ePMP AC with TDMA MU-MIMO  multiple access for  multipoint networks. And  if  it  happen,  it will be  completely other product ( not only GPS sync will be advantage)  in comparison  with UBNT  AC.

2 Possibilities:

1. You are a cambium employee and get fired now as you are whistleblowing

2. You have a realy big glass sphere at home 

I would like you get fired ;-)

40

2 ste

I am not Cambium employee. Also I am  not somebody's  employee at all. So I canot be fired  :-).

41

roanwifi wrote:

Are we talking about same ubiquity m5 airmax? rocket m5? BY FAR I have never seen such figures even on the most ideal signal/better noise level. For PtP < 8k-9k. For PtMP < 2-3k. And these figures are very dependant on distance, due the ACK protocol, if distance increases, the airtime silences increases and  performance decrease.

_______________

We are talking about ubnt  rocket M5 airmax. Without packets aggregation it really has  5K pps. For example, 802.11a gear ( Ubnt Nanostation 5,   AR 2315 Atheros MIPS 4KC, CPU 180MHz)   has only 4K pps.  Look at graph 802.11a  ( no packets aggregation )  Throughput vs packet size. 

Throughput_vs_packet_size.PNG
Throughput_vs_packet_size.PNG1308×1070 6.98 KB

Packets aggregation  of 802.11n  systems   increases max  pps. You may simply measure max pps by testing link  by iperf UDP duplex 64 bytes packet size.  Rocket M5  with 802.11n aggregation passes in lab only  Rx 5-6 Mbps + Tx 5-6  Mbps   duplex  UDP  packet 64 bytes size, that is equal to 22K pps. In  aggregation Off  mode Rocket M5 passes  1.2 +1.2 Mbps duplex  UDP  packet 64 bytes, that is equal to 5k pps.

 Ubnt link pps depends on besides distance also on packet aggregation level , especially in multipoint, where aggregation  falls with increasing number of served CPEs, that causes pps and throughput degradation.

We can see also the same behaviour of  ubnt AC.   ePMP also aggregates UL and DL subframes in multipoint, but it is different  from  ubnt packet aggregation, based on 802.11n protocol, and ePMP has not such throughput degradation as ubnt (N and AC) has due to among other things also pps reduction.

2 Likes
42

I have airMax N sectors doing 3-4k packets with plenty of overhead in a PtMP situation.

43

in your scenario 3-4kpps, what is the distance (min/max/med) of the CPEs to the AP?

44

Doing simple (and may be innacurate) maths based on https://sarwiki.informatik.hu-berlin.de/Packet_transmission_time_in_802.11, this is the expected performance on an ideal non TDD system (those which ACK each packet like airmax, they claim to be tdd but tdd is not what they do) and simplified for 1 CPE, for the best modulation MCS15, counting only DL data, no UL user layer (eg. tcp acks), etc

For example, for the mean download packet size (~800 bytes), and  supposing same SIFS/DIFS times as 802.11a (for n they may be some uS lower):

Tx AP -> SM 800 bytes = 97 uS

TX SM -> AP ack = 57 uS

this is for distance=0 and continous transmit, then doing simple maths, if the raw transmit rate is 150 Mbit/s => the "real" transmit rate (counting, sifs, difs, etc, not the ip overhead) (96 uS for 800 bytes) =>

800*8bit/pkt*1pkt/(96+57)uS*1e6uS/1S*1Mbit/1e6bit => ~42 Mbit/s

if d=1km => add ~7us travel time => 40 Mbit/s

if d=10 km => add ~70 uS rtt => 28 Mbit/s !

if d=20 km => add ~140 uS rtt => 19 Mbit/s !!

these are IDEAL figures for 1 CPE, at max modulation both tx/ack, without retransmisions and WITHOUT the uplink data (eg.if the connection is TCP, there are TCP ACK messages coming from SM to AP which are not being considered right now; if so, the figures go down more quickly)...

If you count UL data (eg 1 tcp ack 64 bytes each DL data packet) => sum another data ul (for 64 bytes=>57uS) and ul ack (57uS) =>

DL + UL for d=0 => (800+64)*8bit/pkt*1pkt(96+57 (for dl) + 57 + 57 (for ul))*1e6Us/1S*1Mbit/1e6bit= 26 Mbit/s...

idem for d=1 km => 24 Mbit/s

for d=10 km => 17 Mbit/s

for d=20 km => 12 Mbit/s

these figures can improve with 802.11n aggregation (aggregation improves a lot PtP scenario but is of low impact on PtMP). As you can se the only way to increase the bandwidth on these equipment is to increase modulation ( 802.11ac) or increase the bandwidth (80, 160, etc.. mhz). Please note that when anybody say "I'm using 20 MHz channel", really you are using 40 MHz of spectrum due to both polarizations.. So when using 80 MHz really it means 160 MHz!!!

For any real TDD system, the performance doesn't degradate such a lot with distance, it is almost despreciable.

Regards

Antonio

1 Like