Medusa optimization advice

Hello,

We've put up our first Medusa sector and I'm hoping for some optimization advice. We're seeing in link test multiple VC results in group 4 and 5 being mostly full but only seeing roughly ~1.5 throughput at max. 100% frame utilization quite a bit. This AP was a replacement for a regular 450 AP at roughly same heading/mount point. Good disparity of customers, clean channel, 9 contention slots, 85% downlink 15% uplink.

Slot Grouping

medua-slot-grouping.jpg

 

medusa-utilization-24hr.png

medusa-throughput-24hr.png

Anything else to really check? All 450 SM's all running 15.1. AP has sync from Timing Port.

1 Like

Hi Timothy,

Your Medusa set up looks pretty good to me. It's probably worth me clarifying a couple of points though as they have caused confusion before.

When you run the link test to multiple VCs the table is headed 'Group Size' , there are not only 7 possible groups but many, many possible combinations of VCs that will form a group with between 1 & 7 members. The members of the group will change every TDD depending on the traffic which is waiting to be sent to VCs in the sector. In the link test all VCs will have data to be sent and the biggest groups possible will be formed each TDD but with different VCs. Your results show Medusa is able to form groups of 4 or 5 with all your VCs so the potential throughput improvement should be much larger than the 1.5 you have seen. Try the same link test but with the MU-MIMO radio button set to disable and you should get an idea of the throughput improvement the MU-MIMO spatial multiplexing is giving you (I would be interested if you want to share the 2 results total throughput).

See also this thread, which in turn has link to a previous one !

grouping thread

OK, next the 'only 1.5x improvement and 100% utilisation' points. Looking at the graphs, at no time is the throughput reaching a plateau. So I think the only limiting factor here is the bandwidth your users are demanding, the test mentioned above may be more useful until they start streaming more netflix ....

The 100% utilisation metric is one I don't remember seeing mentioned in the forums recently but can definetly be confusing to new 450m users as it should NOT be used in the same way as 450/450i. We have some new metrics coming out for Medusa (in release 15.1.1, I hope) that will be more useful, let me quote a couple of paragraphs out of the requirements document we have for this as it's relevant here, sorry it's a bit long - I'll do a TLDR afterwards.

"

Utilisation metrics
The existing Frame Utilisation metric in the PMP450 AP provides an indication as to how close a sector is to the maximum capacity based on the AP traffic in the sector over the past 15 minutes. Some service providers who perform dynamic traffic shaping use regular readings of the Frame Utilisation metric as one input to their algorithms that determine their traffic shaping settings. For example, one WISP applies dynamic traffic shaping with the objective of limiting PMP450 Frame Utilisation to a maximum of 75%.
On a PMP450m with MUMIMO enabled, the Frame Utilisation metric shows similar results to that reported by a PMP450 or PMP450i AP under identical traffic conditions. This occurs even when the traffic is delivered to many spatially diverse users where the PMP450m had the potential to exploit spatial multiplexing. The reason a PMP450m reports similar results to that of a PMP450 or PMP450i AP occurs because there is minimal spatial multiplexing gain until the Frame Utilisation approaches 100%. The spatial multiplexing gain of PMP450m only kicks in when the Frame Utilisation is near or at 100%. Consequently the Frame Utilisation metric is not a useful indicator of how close the recent traffic load is to the MUMIMO capacity of a sector.
To address the limitations of the Frame Utilisation metric with PMP450m, the MUMIMO Utilisation metric is being added. This provides an indication of how close recent traffic is to the MUMIMO capacity of the sector. The MUMIMO Utilisation metric estimates the proportion of downlink airtime resources being consumed based on recent traffic assuming data in the VC queues was available early so that optimal MUMIMO grouping could be performed. The timing advance required appears well within that accommodated by latency tolerant traffic such as streamed media."

So, unless you are getting near 100% utilisation then there is no need for the MU-MIMO gains to be used and the algorithms will work as before targeting individual VCs  as this will give the absolute best throughput to that VC. Once you start getting near 100% utilisation MU-MIMO gains start to kick in and you will see the improvement over your previous 450 sector system operation.

Hope that helps, let me know if any points require further clarification. Congratulations on your first 450m install, it looks like a good one.

best regards,

Andy

6 Likes