Reply
Ubiquiti Employee
Posts: 6,290
Registered: ‎05-13-2009
Kudos: 1944
Solutions: 191

Re: airMax - aggregation - priority

What do you mean by channel width? Give me an example comparing 5mhz and 20mhz.


Lower rate = longer transmit time. If comparing 5Mhz to 20Mhz in the same link conditions, aggregate size for 5Mhz channel width would be 4 times smaller than for 20MHz.

-Edmundas
Established Member
Posts: 1,005
Registered: ‎09-29-2009
Kudos: 66
Solutions: 1

Re: airMax - aggregation - priority

Lower rate = longer transmit time. If comparing 5Mhz to 20Mhz in the same link conditions, aggregate size for 5Mhz channel width would be 4 times smaller than for 20MHz.

-Edmundas


That's what I thought you were getting at. So what do my results mean to you when I tell you that I use 5mhz chw across my entire network?
Rural Crossing
Internet & Community Networks
Northeast Indiana; Northwest Ohio; Lower Michigan
Ubiquiti Employee
Posts: 6,290
Registered: ‎05-13-2009
Kudos: 1944
Solutions: 191

Re: airMax - aggregation - priority

It depended on the link. That 900mhz bh I *****ed about for a year doubled in throughput when I increased bytes to 65000 on the AP side and lowered bytes to 3839 on the sta. I kept frames @ 32. The reason I lowered the sta side is because I sell asym access and most of the outbound from subs should be acks.

I lowered the bytes to 32000 on APs with subs in nLOS or NLOS areas and set all subs to 3839.

Still fooling with combinations...spent the last 2 days doing service calls because of dish spins. The pic shows how windy it gets around here.


That's what I thought you were getting at. So what do my results mean to you when I tell you that I use 5mhz chw across my entire network?


Looks like your Stations does have problems with larger frames and lowering aggregation size increased their performance.

-Edmundas
Established Member
Posts: 1,077
Registered: ‎09-17-2010
Kudos: 184
Solutions: 4

Re: airMax - aggregation - priority

I'm dying here...what setting did you use???


I've found a better value myself in testing 5mhz.

My default download speedtest would get around 5.79 Mbps, with pawpaw value of 32,000; I would get 7.39 Mbps on speedtest now (quite an improvement), but I've found that using the old multiple bit trick that programmers love (power of 2), that the value of 16,384 would score me 8.57 Mbps on speedtest. Seems to be the best value I've found for the AP when dealing 5MHz channel widths.

This was using an Rocket M5 GPS (AP) to an AirGrid M5 (CPE). I haven't seen what it does to the dual-polarity radio yet (like a NanoStation M5), but I'm experimenting with CPE values now to see which scores the best return throughput to the AP.
Established Member
Posts: 1,077
Registered: ‎09-17-2010
Kudos: 184
Solutions: 4

Re: airMax - aggregation - priority

The best value I've found for the CPE in testing has been the lowest I could set, 2304, this result in a 1 Mbps boost in upload speed as well better response to the web GUI on the CPE and better latency (lower pings) for CPE with rather poor signal.

I'll let these run for a few days to see if any of my power users notice (for better or worse), but it all holds out, these seem to be better default values for the 5Mhz channel width as far as throughput goes when using airMax.
Established Member
Posts: 1,077
Registered: ‎09-17-2010
Kudos: 184
Solutions: 4

Re: airMax - aggregation - priority

Just tested some dual-polarity radios, NanoStation M5 and NanoStation L M900, setting the AP made just a little difference, maybe 0.5 Mbps of better download bandwidth, so really I could that to have really no effect since speedtest can vary by that much with the default settings. As far as setting the CPE, the lowest value 2304 actually hinders the upload of the CPE, but a value of 4096 returns to the same speed as the default 50,000 value. So basically anything better than 4096 had no real benefit.
I'm going to let that site run at these values for a while and see what my power users think about it.
So far, the benefit seems to be to single polarity devices in 5MHz channel width than the dual-polarity radios. But if these values work to improve one without hurting the other, it's certainly something to keep a watch on for the future.
SuperUser
Posts: 21,761
Registered: ‎11-20-2011
Kudos: 7894
Solutions: 233

Re: airMax - aggregation - priority

Just tested some dual-polarity radios, NanoStation M5 and NanoStation L M900, setting the AP made just a little difference, maybe 0.5 Mbps of better download bandwidth, so really I could that to have really no effect since speedtest can vary by that much with the default settings. As far as setting the CPE, the lowest value 2304 actually hinders the upload of the CPE, but a value of 4096 returns to the same speed as the default 50,000 value. So basically anything better than 4096 had no real benefit.

I'm going to let that site run at these values for a while and see what my power users think about it.

So far, the benefit seems to be to single polarity devices in 5MHz channel width than the dual-polarity radios. But if these values work to improve one without hurting the other, it's certainly something to keep a watch on for the future.


Try multiples of the wireless mtu, which should be...

(logs into a 5.5 locom5)

XM.v5.5# cat board.inc | grep mtu
$eth_mac1_max_mtu=9192;
$eth_mac2_max_mtu=2024;

Post your results.


isp builder | linux sorcerer | datacenter automation conjurer | blog: blog.engineered.online
link to our slack channel on the blog
Established Member
Posts: 1,077
Registered: ‎09-17-2010
Kudos: 184
Solutions: 4

Re: airMax - aggregation - priority

Try multiples of the wireless mtu, which should be...
(logs into a 5.5 locom5)
XM.v5.5# cat board.inc | grep mtu
$eth_mac1_max_mtu=9192;
$eth_mac2_max_mtu=2024;
Post your results.

I had tried a lot of values in between before, started with the lowest allowed value of 2304, then did a try with 4096, 8192, 16384, 32768 before I posted and nothing seemed to improve the upload speed of the CPE.
This time, I did a multiple of the 2024 since it matched my loco exactly, started with a base. These test were run at 3am, so literally no traffic was across the AP except maybe DHCP traffic if any.
CPE (5MHz)
Base Aggregation 4096 -> 5.22 Mbps average
Comparative:
Aggregation 2304 -> 4.26 Mbps average
Aggregation 3036 -> 4.27 Mbps average
Aggregation 4048 -> 4.92 Mbps average
Aggregation 8096 -> 5.17 Mbps average
Aggregation 16192 -> 4.69 Mbps average
Aggregation 32384 -> 4.69 Mbps average
Aggregation 50000 -> 5.30 Mbps average
The reason I used the multiple of 2 for my testing (previous posting) was that I saw the maximum was 65535 (counting 0 as a number, that's 65536 possible) from my programming days, as this number that was representing a 16-bit binary number.
So you may ask, if there isn't much difference between 4096 and 50,000, why use a lower number? Well the lower number is smaller packets, which results in a more snappy data exchange between the AP and CPE. Side-effects, the web gui on the CPE is faster (the main window with all it's java now comes up as fast as it does for a radio that is on the physical lan with myself like the AP for example) and I'm hoping that this will make a difference between "stalls" on the CPE from heavy traffic situations, but I won't know until some time passes and everyone is actually keeping the AP busy during peak times.
SuperUser
Posts: 21,761
Registered: ‎11-20-2011
Kudos: 7894
Solutions: 233

Re: airMax - aggregation - priority

Just FYI, I didn't notice any noticeable difference in a PtP scenario between aggregation at 2024, and aggregation at 64768 (pure multiple of 2024).


isp builder | linux sorcerer | datacenter automation conjurer | blog: blog.engineered.online
link to our slack channel on the blog
Established Member
Posts: 1,005
Registered: ‎09-29-2009
Kudos: 66
Solutions: 1

Re: airMax - aggregation - priority

Try lowering frames as well.
Rural Crossing
Internet & Community Networks
Northeast Indiana; Northwest Ohio; Lower Michigan
Member
Posts: 243
Registered: ‎11-15-2008
Kudos: 37

Re: airMax - aggregation - priority

Yes, increasing aggregation size actually hard to believe that could make any difference due to the limitation inside wireless driver. Smaller aggregation size would make sense.
4ms is a time needed to process aggregate (how driver determines aggregate size) in wireless driver, it's not a airtime given for station or a created latency what you could see with various test tools (i.e. ping). Also I don't think that I could give you more detailed explanation, as I'm not the wireless driver developer and don't know how exactly everything is processed there inside.
-Edmundas

Are you trying to say that the wireless driver waits up to 4ms for enough packets to accumulate to fill the current aggregate, whatever size it may be?
This might explain how smaller max aggregate sizes result in snappier response times. Especially if the driver is trying to fill to the max aggregate size instead of the current aggregate size. (Current aggregate size would be based on link conditions)
SuperUser
Posts: 21,761
Registered: ‎11-20-2011
Kudos: 7894
Solutions: 233

Re: airMax - aggregation - priority

Are you trying to say that the wireless driver waits up to 4ms for enough packets to accumulate to fill the current aggregate, whatever size it may be?

This might explain how smaller max aggregate sizes result in snappier response times. Especially if the driver is trying to fill to the max aggregate size instead of the current aggregate size. (Current aggregate size would be based on link conditions)


... "yes".

There is a tradeoff between latency and throughput in that regard though.


isp builder | linux sorcerer | datacenter automation conjurer | blog: blog.engineered.online
link to our slack channel on the blog
Emerging Member
Posts: 60
Registered: ‎04-16-2008

Re: airMax - aggregation - priority

In ptp scenario, high noise, for aircam video h264, what is the suggest value of packet and size aggregation ?

I tried to set off aggregation and also if the speed test is around 10 + 10mb full duplex, frame rate of the camera go down from 25fps to 4 !

Reply