We've been doing WISP stuff for almost a decade now. We started out with a lot of WiFi-based stuff, the original networks were SkyPilot 5.8 GHz 802.11a and Smart Bridges 802.11b (WISP that we purchased). We got into PMP-100 on 900 MHz, upgraded the 802.11b stuff to 802.11g Tranzeo equipment, bought an ISP with 2.4 GHz PMP-100 and 802.11g Ubiquiti gear, and did a couple hundred PMP-320 WiMAX customers. We're now rolling out PMP-450 everywhere we can, and I have to say, this stuff rocks.
Our network is constantly changing and we currently support everything I listed above except the Ubiquiti gear (but it works the same as Tranzeo anyway). Our marketing is designed for plans that basically work across the network.
Up to 5/1 Mbps with 50 GB or 100 GB
Up to 10/2 Mbps with 250 GB
"Max" - fastest speed on that platform, with 350 GB or 500 GB. On PMP-450 this is 25/5 Mbps.
Since this is the PMP-450 forum, I'll focus on how we provision that gear.
Unlike Eric, we do not use any third-party traffic shaping or packet inspection. The first WISP we bought made use of a NetEq, and another WISP I consult for does as well. They're pretty handy boxes, if you understand their limitations, configure them properly, and can fit it in with your network and budget. We use PPPoE for all of our customer connections, so that kind of screws up that tech anyway. Instead, we use two different mechanisms:
The first is bandwidth quotas, which let customers self-regulate their consumption. Plans are really priced according to usage. We did some analysis and concluded that the access speed doesn't matter to us as much as it used to. Of course, if everyone only uses their service at peak time this is doesn't work well, but it cuts down on serial heavy usage and is a nice source of income. My employer doesn't pay "that much" for Internet bandwidth, but the WISP I do consulting for is in the middle of nowhere and Internet is EXTREMELY expensive - they were losing money because their Internet bills were exceeding the revenue from subscribers. Implementing packages with quotas helped them deal with this issue immediately and get heavy users to pay their share (incidentally, also done via a NetEq - I guess the new software is pretty cool)
The second mechanism we use is the queuing features built into the PMP-450 gear itself. I highly recommend digging into the manual and figuring out the QoS section. This gear will provide a different rate to a subscriber in a burst of activity than on longer transfers. The basic idea is that this promotes responsive web-browsing and other intermittent applications, but keeps under control stuff like streaming video and file downloads.
On PMP-450, our actual profiles are set up like this:
5/1 Mbps package:
- DL burst size: 24 MBytes
- DL burst rate: 6 Mbps
- DL sustained rate: 3.5 Mbps
- UL burst size: 10 MBytes
- UL burst rate: 1.5 Mbps
- UL sustained rate: 768 Kbps
10/2 Mbps package:
- DL burst size: 48 MBytes
- DL burst rate: 12 Mbps
- DL sustained rate: 9 Mbps
- UL burst size: 20 MBytes
- UL burst rate: 2.5 Mbps
- UL sustained rate: 1.5 Mbps
25/5 Mbps package:
- DL burst size: 96 MBytes
- DL burst rate: 30 Mbps
- DL sustained rate: 20 Mbps
- UL burst size: 20 MBytes
- UL burst rate: 6.5 Mbps
- UL sustained rate: 4 mbps
The cool part about this approach is that we deliver an experience like a faster connection, but it actually uses less bandwidth long-term. Speed test results actually match the "burst rates" above, so anybody on the 10 Mbps plan actually sees 12 Mbps on a test. Cool huh? For the most part this applies to the whole experience. But certain apps, like NetFlix, are bandwidth-hungry and will just keep scaling up to use the ENTIRE pipe. Obviously this is an issue. Boxes like NetEQ and Sandvine DPI screw with the traffic of these apps enough to keep them under control. This is a low-tech approach that gives the same results while being application-agnostic.
We don't have any major congestion problems except on a couple APs with a lot of heavy users (one AP faces a subdivision with >50 subscribers on 10 and 25 Mbps plans). In response to congestion we actually clamped the sustained rate - 10 Mbps profile down from 9 to 7 Mbps sustained, 25 Mbps profile down from 20 to 10 Mbps sustained - and this has resolved complaints until we can put up the additional hardware to relieve congestion.
Our PMP-450 links are pretty well engineered. 95% of them run 8X/8X, good fresnel clearance, reflectors out the ass. Sketchy links go on another platform like 900 MHz PMP-100. Running poor links is a MAJOR hinderance to AP capacity, regardless of platform. Keeping up with AP capacity is probably the biggest challenge, but it's a lot easier with PMP-450.