With the release of 15.1.3 and the support of 30 & 40MHz channel widths, I wondered if anyone had moved to the new bandwidths because they were fully maxed out on 20MHz ? How is 30/40 MHz working for you ?
Which leads on to the more general question of what throughput 450m users are seeing on a sector. If you are monitoring throughput, do you ever see it plateau ?
Do the 'Sector Utilisation' and 'MU-MIMO Utilisation' percentages (under Statistics/Frame Utilisation) ever get near 100% ?
How does your max throughput compare to the 'Link test to multiple VCs' test ?
Please share if you can, along with some background (e.g. bandwidth, sw version, number of VCs etc)
Happy to help Andy. We moved from 20 MHz to 30 MHz. Posting a 2 week view of our graphs, and we upgraded this AP about 1 week ago. Real world data. Shows fewer RF discards but also less throughput. We went from 15.1.1 to 15.1.3 B7.
Thank you, this is very interesting.
If you could let me have an engineering diags (just put IP_ADDRESS_OF_AP/engineering.cgi into a browser address bar and save the resulting page) from this unit that would be great ( firstname.lastname@example.org ) especially captured during a ‘busy’ time.
I’m not quite sure what your ‘RF Utilization’ is a measure of and how it might relate to sector/frame/mu-mimo/su-mimo utilization.
Less RF discards is good.
Ethernet traffic seem to be peaking at 80M for the last 4 days, which matches the 16th & 17th when the SM count was the same. This is somewhat at the mercy of your users, maybe it will get busier again this weekend.
From the link test result, I’m guessing you could get 2-4x MU-MIMO gain, so there should be plenty of overhead in the sector as yet unused.
Thanks again. Andy.
This is a good example of a sector that, while busy, may still have some room to get better at the busy hour... i.e. have additional capacity to give.
If you don't agree (i.e. you believe the sector is full and not capable of increased capacity beyond what you're showing), let us know.
Also, looking for other examples too, please post them here.
We are currently trying out our first deployed PMP450m. We mainly use PMP450i APs at the moment but also have a few sites with the older PMP450 AP. The 450m was installed to augment and possibly replace some existing 450i APs that were full and that had some overlapping coverage. After running this for a couple of weeks we can offer the following perspectives:
1: The RF performance of the 450m is very good and the client SNRs are generally the same or better than previously. Most clients have 6x modulation rates with signal levels in the mid to high 60's. We are using 30MHz channels and are running the latest released firmware version 15.1.3 with a 75:25 download / upload split.
2: We have currently moved about 100 clients onto the 450m, most of these have packages that allow them to burst to whatever speed is available to them. There is also a good spatial spread of clients, in terms of the azimuth, from the centre point of the coverage area.
3: When the 450m is at max utilization in the late evening we are seeing the download frame utilization at about 90% with five minute average download throughput of about 140mbps. This is about double the capacity of what we typically get using the 450i AP.
So, in summary, we are currently seeing approximately double the throughput in the real world between the 450m and the 450i AP on 30MHz channels. Possibly the increase in capacity should be higher but this is what we are seeing.
Thanks for your post, I really glad it seems to be working well so far.
With the 450m the MU-MIMO benefit only starts to kick in as the frame utilization gets to around 80-90%, before this the scheduling will use SU-MIMO as that is perfectly capable of handling the traffic. As more MU-MIMO is utilized then the frame utilization will plateau around 90%.
So on a 450m it is more important to watch 'Sector Utilization' which will be made up of 'MU-MIMO Utilization' and 'SU-MIMO Utilization'. As the load increases you should see the the MU-MIMO start to kick in (along with the 'Multiplexing Gain' value going up).
140Mbps is great but I suspect you can get more than that if the users requested it, if you want a more detailed explanation and break down of your specific sector then feel free to send me an enginnering.cgi from a busy period and we can discuss it privately or report it back here.
Thanks again, you're currently in 1st place for Mbps on this thread !
We graph the Frame Utilization, like we do for regular 450/450i AP's. I see your last post recommends we graph Sector Utilization. However, i'm not sure one can see if the AP is overloaded by just looking at that. There needs to be an easier way to tell that the AP is overloaded for one user who may be experiencing slow service. Right now, I believe the only way to tell is to find out which VC the SM is using and watch the Spatial Utilization table live.
The problem is, if a customer says, "service was slow at xyz time", we have no way to confirm or deny if the bottleneck was at the AP or somewhere else. Here is another Medusa graph on our network we upgraded about 30 days ago. This is still running 15.1.3 Beta 3, 30MHz channel. I will send you a engineering.cgi later on.
Thanks for adding more data plots.
Along with the Spatial Utilisation you could monitor the data available in the Statistics/Data VC page.
This shows the packets going to each VC, along with discards,errors and queue overflow. These last three could be used to identify a struggling VC.
We are happy to help diagnose things if you get a capture, where there might be a problem.
Also if you have ideas about additional stats that would be helpful to you then please let us know. There's lots of stuff we can monitor but getting it displayed/logged in the best way for customers obviously works best with your feedback rather than our engineers best guess (not a complete guess we have worked with beta customers to get where we are today!) as we don't run networks ourselves.
This is a graph showing throughput and Frame vs MU-MIMO Utilization.
It is from our test chamber, so has an optimal spacing of SMs.
It was captured quite a while ago I think we have improved the throughput since then (TCP should be >550Mbps without any issue).
If anyone is interested in a more up to date version let me know (along with what stats you would like to see).
Assuming that test was only using a 20MHz channel?
Yes, 20MHz, I really should have mentioned that !
Yes, with the current release, we can repeatably demonstrate about 550 Mbps of downlink traffic.
Here's some images of our experience over the last ~60 hours (timeframes are all reasonably synchronised):
Sector Utilisation via medussa OIDsClassic frame utilisationThroughput
That is an excellent set of graphs showing the relationship between the different utilization metrics (old & new) and throughput.
We can see that as the 'old' frame utilization goes >80% that the MU-MIMO kicks in. The new sector utilization metric still shows spare capacity only hitting around 60% as the throughput hits 141Mbps. Lovely.
If you would like a free T-Shirt for your efforts please contact Ray Savich and point him to this post !
In the meantime I shall give you a 'kudo' :)
What are the OID's you are grabing for the MU-MIMO information?
Graph from last night (132 subs, 20 MHz wide channel, 2.5ms frame size)
The new best I think in terms of max DL.
lightnet_barry just needs one or two more netflix streams to take back the lead though :)
This post describes the OID's we would like to look at.
monitoring utilization 450m
Look forward to your input !
I don't really need to go to 40mhz channels in the winter as most of the customers in this area are cottages but! In a 10mhz channel with the new 5ms timing I'm seeing great throughput.
20 MHz channel
Channel with interference of -78
This is our best result (modulations, signals etc...)