The importance of contention slots!

Recently we've been experiencing some issues with some of our more heavily loaded PMP450 AP's dropping packets and having increased latency to almost all clients. We also had some clients reporting speed issues, usually buffering with Netflix. We noticed that frame utilization on both the downlink and uplink would hit at or near 100% frequently. Here's an example of the BW I was pushing and the frame utilization on a heavily loaded AP. This AP has 60 clients on it, it uses a 20mhz channel width, and it's frame settings were 80/20, and 2 contention slots, 0 for broadcast repeat count:

You can see at peak, before changes, I was only getting a max of around 41mbps on the downlink, and 2.8mbps on the uplink.

You can see frequently during the day, and more so during peak hours (6pm to 11pm) that frame utilization would hit at or near 100%. The big dip you see around 6-7pm on Monday is when I did some peak hours contention slot experimentation. Monday at midnight and on from there represents the new settings I finalized on. Notice how much lower it is in general and the downlink and uplink utilization matches each other closely with very little gap in utilization.

We first tried changing the frame ratio from an 80/20 split to a 70/30 split... this helped a little bit, but we were still seeing packet loss and latency to clients. We then read the "PMP 450 Contention Slots" white paper from Cambium (see attached). After reading the paper, I realized that perhaps by having too low number of contention slots, I was perhaps 'starving' the clients of upload slots. While we don't have a lot of upload users or VoIP, everyone downloading requires a bit of uplink BW. We have a lot of users that use Netflix, which uses TCP... which requires constant acknowledgments via uplink. if you starve the uplink, then you can actually decrease download BW. You don't want to have too many contention slots as each one allocated uses more BW... but it seems using too few is even worse overall for performance. I decided to play around with more contention slots, and keep the 70/30 frame ratio split. After quite a bit of testing, I settled on 5 contention slots for this AP.

You can see there was quite an increase in peak download and upload BW, and all the latency and dropped packets have disappeared:

Peak download BW has increased from 41.6mbps to 54.8, and upload BW has increased from 2.8mbps to 3.6mbps. While this increase in uplink bandwidth doesn't seem huge, it does represent a huge increase in the number of uplink acknowledgements available for streaming apps like Netflix. A less then 1mbps increase in upload has increased download bandwidth available by over 10mbps or over 20% more bandwidth. Lastly, frame utilization has decreased and is now much more stable throughout the day in during peak times and we appear to have more room for growth now.

I went ahead and replicated these settings on all my PMP450 AP's that were having similar issues and all of them have improved significantly.

The takeaway here is monitor your frame utilization and adjust your contention slots accordingly!

11 Likes

Thanks Eric,

That is really useful and informative for everybody.

Unfortunately it looks like it means we've lost the opportunity to upsell you a 450m for that sector though ....

best regards,

Andy.

4 Likes

Although I havent had issues (yet) I also will be moving to 5 slots on all my AP's. Thanks for taking the time to write this as it helps valadate that we are doing the right thing!

Ben

1 Like

I'm curious how the new 'dynamic' contention slot option in the upcoming 450 firmware will work. I'm also curious about how contention slots are used on 450m. BTW we're getting ready to light up our first 450m 5ghz AP! We're also eagerly awaiting the 450m on 3.XGHz... which we'll be deploying quite a bit of!

2 Likes

Hi Eric,

I'm looking forward to your feedback on the 450m.

Contention slots are used in the same way on 450m. Something we don't usually mention is that we can receive data from multiple VCs in the same contention slot with 450m. It all depends on the spatial frequency of the VCs etc and the beam pattern chosen for the contention period. We try to optimise the opportunity for this but as it can't be easily measured or proved and will be changing all the time it's not something we shout about too much. It will be as good as non-450m behaviour and potentially better (e.g. in our test setups with >230 SMs I can see 3 or 4 SMs data in a single contention slot).

The 'dynamic' contention slot option should work great, I've not had the chance to try it yet though.

There is also UL-MUMIMO to look forward to, which we are testing now. This is looking really good even helping on downlink stats too, especially improving TCP throughput to a group of SMs for example. This has bumped a downlink TCP test to 7 SMs from >550Mbps to >630Mbps as all the handshaking (both of the radio and the TCP itself) is not limited by the UL capacity (In a nice lab environment at 20MHz).

Andy.

4 Likes

@Cambium_AndyR wrote:

There is also UL-MUMIMO to look forward to, which we are testing now. This is looking really good even helping on downlink stats too, especially improving TCP throughput to a group of SMs for example. This has bumped a downlink TCP test to 7 SMs from >550Mbps to >630Mbps as all the handshaking (both of the radio and the TCP itself) is not limited by the UL capacity (In a nice lab environment at 20MHz).

Andy.


WOW... that's over 30bpHz!!!! That's incredible, even for a lab environment!!! I think that's the highest bpHz I've ever heard of in an outdoor radio. Amazing, industry leading tech... can't wait to put it to work in the field!!

2 Likes

@Eric Ozrelic wrote:

 We're also eagerly awaiting the 450m on 3.XGHz... which we'll be deploying quite a bit of!


Not sure how much I can mention about this, but the prototypes are also in the chamber here and performing very nicely (grouping well with good modulation etc, even before we've started tweaking things)

Andy.

4 Likes

@Eric Ozrelic

WOW... that's over 30bpHz!!!! That's incredible, even for a lab environment!!! I think that's the highest bpHz I've ever heard of in an outdoor radio. Amazing, industry leading tech... can't wait to put it to work in the field!!


>30b/Hz is misleading because the clients require spatially separated streams, the clients must be spaced right to take advantage of the separate streams from the phased array. With LDPC, but without MU-MIMO, the Ubiquiti LTU product will have a higher raw spectral efficiency than PMP450, albiet without this nice feature...