Basic Tests After the Install: Proxim Tsunami MP.11 & Latency

     After the adventures of Saturday, I have been able to take some time to learn more about tweaking the Feed and the Center. Located at the Feed, I have a Proxim MP.11 Subscriber Unit and at the center, I have a MP.11 Base Station (also known as an SU and BSU). Most of the clients will end up using a Residential Subscriber Unit (RSU) because of a) cost and b) lack of need for the one feature that differentiates it from the SU - unlimited MAC address association. The Center is broadcasting using a 15.5 dBi omni directional antenna about 55 ft in the air and the Feed is transmitting using a 15 dBi directional patch antenna (with a beam width of about 30 deg.) located about 40 ft. off the ground. One of the potential clients of the system asked about network latency... I hadn't actually put much thought into it until the question was asked! How will the environmental and setup factors effect network latency? This is especially important as the system grows. If I have drastic latency issues right now, I can only imagine them getting worse as more users come online.
     To begin simple latency tests, I took a workstation behind the feed-node and pinged the center-node. The results look a lot like:

As you can see, there may be wild variances in the round-trip time, but for the most part, the feed to node round-trip time is around 6ms. At first I was quite a bit concerned about the fluctuations - in fact, I still am concerned, but it is a reality of the OS on the equipment and the environment. Another thing that one should notice is that it will maintain a round-trip time of 6ms for a few pings and then jump to some double or triple digit number. In my estimation, this is partly due to flex of both of the antennas in the wind and reflection issues. One thing I have been reading about is background noise when using an omni. Eventually, if we got over 20 subscribers (which would mean that we would be setting up a ResNet - and is not the current goal of the project), I would consider sectorizing the center-node (i.e. breaking up the omnidirectional center-node and puting in directional patch antennas that cover a 60-120 deg. area instead of 360 deg.). On top of the environmental issues, a feature of the MP.11 series is SU/RSU registration polling periods. It appears that every 5 seconds, the SU/RSU re-registers with the BSU - which would be good reason for latency spiking at what appears to be 5 second intervals. I ran a ping test over night on Saturday and found that the cycle of jumps from 6ms to >6ms was extremely regular - again, every 3rd-7th ping.
     Another setting I toyed with was overall throughput settings - i.e. 11 Mbits/s vs. 5 Mbits/s, etc. I have found on other wireless systems that although a higher bitrate connection can be achieved, the signal strength is low enough that throughput at that bitrate is worse than if we were to simply go to a lower bitrate setting. I played with different combinations and found that the signal between feed and center is good enough that leaving it at 11 Mbits/s doesn't give me worse round-trip ping times than setting the BSU or SU/RSU at a lower bitrate. This may change, however, with the introduction of more SU/BSU's. Luckily, though, as long as the signal is decent, changing these settings won't impact the latency to any great extent.
     The literature is very vague concerning how to reduce latency issues. I think that the first objective of any project is to provide the access and then move on to improving network performance. I am surprised, however, that not much effort seems to be put into this issue - I don't know whether it is because of the apparent connection between signal strength ad latency (it appears that as the signal gets worse, the latency goes up - no surprise there - however, what is the threshold?) or whether it is simply a forgotten issue. I will continue to tackle the problem, however, as I add more end-nodes and see what system parameter changes I can make that will help reduce over-all system related latency.