wendmea uthmen kitten

0 views
Skip to first unread message

Yvone Brem

unread,
Aug 2, 2024, 10:48:03 AM8/2/24
to ebelefin

What is in the nature of Wi-Fi is the request to provide guest Wi-Fi. For some, this is a very dreaded request and usually met with angst and hand wringing. Guest are notorious for having wonky devices that are out of date and unwilling to admit that they need to pony up and purchase a new device that is up to some type of current standards. A great write up about that came from Lee Badman over at @wirednot and if so inclined, he offers a decent insight to the thinking of the Wi-fi administrator who is met with this request.

This seems like a pretty simple step so I boiled down my question to this. Can having Netflix users on your guest Wi-Fi negatively affect the WLAN environment, and in turn, congest the system to the point that mission critical and/or life saving devices can no longer function, in turn leading to loss of life and/or revenue?

Most of this is anecdotal at best, this I will admit. A quick search of Netflix and Wi-Fi will turn up numerous hits about how important these two items are, especially in the world of hospitality. My best suggestion to prove this yourself is get a couple of Wi-Fi architects, engineers, and administrators in a room and just ask the question about how Netflix impacts the corporate WLAN. You better pack a lunch because you are going to be a while.

While heavy downloading can and will have a negative impact on a WLAN environment, my belief is that all operators of guest Wi-Fi networks greatly over-exaggerate the amount of this heavy downloading, and in turn, make changes and insert devices or mechanisms to throttle this activity in an attempt to prevent the negative impact from happening.

My hypothesis is that this very act of trying to limit the impact of these devices is instead creating a greater impact on the overall health of the network, sometimes to the detriment of the corporate wired and wireless network.

The stated purpose of these tests are to discover the impact to both the wired and wireless network a single user has on a WLAN system. As such, I set up a test to measure what would happen as rate limits were applied in different increments. My testing set up is as follows:

All of the graphs you are going to see were pulled from a pfSense firewall that acted as my firewall and all Layer 3 functions on my private VLAN. Also, these graphs are updated about every 1 or 2 seconds so they could update as fast as I wanted them to. Unfortunately, none of my other tools updated that much, so the resolution was much less. This does call into questions about what these graphs are showing, so let me walk you though it before we continue.

The X Axis of the graph is showing the minutes and seconds pulled from the firewall system clock. What you are seeing is not hours and minutes, but minute and seconds of the day. The time of this test was actually 06:13:39, not 1:39 PM. This will also indicate that this graph only shows 2 minutes of time. If you see a graph that starts, peaks, and then falls off all within the display, it means that event took 2 minutes or less of actual clock time.

Speaking of the table on the right, this shows the current LAN traffic and the client IP involved. Depending on the direction of the traffic, upload or download, it will show the current speed of that traffic. This table is not historical, it only shows active traffic. If there is no traffic, that table will be blank. In those cases, I have removed the table in some instances so the graph can display bigger.

As you can see, the Y Axis has adjusted to match the rate limit, the time to download the file is now longer, and as proof of the bug I mentioned earlier on the Y Axis, you can see the actual rate that the client is getting at that second on the right side, not the smaller number on the Y Axis. However, we still get to see the spikes in the throughput that we like to see in a well performing network. This is critical later, so remember this.

As you can see, the graph gets really flat, and the length of time to download movies gets longer. Due to the linear path of the testing, I felt like this was a waste of my time, and my graph of the data backs me up. At least I think it backs me up:

For the download testing, the premise was end users were going to ask their device to pull a full video file from a remote server to store on their device so they could view that content at a later date. You remember, the underground cave scenario.

As you look through the three graphs, you can see the time move as I watched the movie. The other thing I found interesting that I will point out on the first graph is what I call the setup period. This is where the device is pulling the first bit of content down to the device to buffer. The set up data and then the first pull of data was interesting, and then it settled down to a routine pattern. Every 20 seconds or so, as the buffer on the device was depleted by playing the video on the screen, the device would go back to the well, so to speak, to top of.

YouTube streaming is such a different beast than Netflix. Content can be a mix of professionally edited, HD videos and amateur cell phone clips. The graph above shows a random video that was suggested by YouTube. More spikes, a little bit wider, and not as much data transferred per buffer refill. 5 Mbps compared to the 50 to 80 Mbps seen in Netflix streaming with no rate limit. Less data transferred (rate x time) means more time going back to the resource.

At 2.5 Mbps of rate limit, we see the same thing as Netflix. An almost constant demand on the time resource, all the while still not pulling a ton of data on each request. Add in a second user streaming Youtube and you can imagine how those white spaces representing time just go away, meaning that a third user would then start to aggregate the speed needed to serve the three clients.

Now, at your house or Small to Medium size Business (SMB), this may not be of a great concern. What I need you to do next is to extrapolate this with me to a much larger client count that you may see at large public venues or hospitality or healthcare.

From Certified Wireless Network Professionals (CWNP) Certified Wireless Network Administrator (CWNA) training, and demonstrated by Joel Crane at Wireless LAN Professionals Conference (WLPC) in 2018, Wi-Fi is still half duplex.

Thanks to Peter for allowing me to use this chart. If you have never seen it, this is what is required for any one device to send a frame on a channel. The AP is treated as a device on the channel, so it has to play this game as well. Remember my comment about the way AT&T TV has their buffering running from earlier?

Between 802.11 inherent limits, remote server capabilities, half duplex, and DCF, there are already mechanisms in place that keeps end users from claiming all of the resource and not allowing that critical piece of life saving equipment to get their one frame through.

This seems a little redundant at this point of my blog post/white paper, but I am trying to follow the scientific method! This is something that I have been working on for some months now, and the initial results were so intriguing (rate limiting a streaming device below 10 Mbps uses almost as much time on the resource as a download) that I actually did a presentation at WLPC 2019 around this same topic, even using some of the same graphs.

I would like to comment or suggested idea, that we were are able to reduce the aggregation demand of the internet pipe for one of our customer and their specific use, by applying a daily data quota to each user (not each device, they were allowed to use multiple devices, but quota was shared across user devices)

Can you connect your Mac to a different network to help rule out the issue being related to your network connection? This would be trying an Ethernet connection if available, or just a different Wi-Fi network connection: Connect to the internet with your Mac

My observation is that the Video Quality seems to dip in and out and at one time I thought it was the network. Running the Activity Monitor and watching Safari network activity and throughput suggested that every time the video deteriorates the network went slow - not the network but Mac's access. I thought I'd found it by removing a PROXIE setting and selecting Auto Proxy Discovery. Then, instead of allowing the Location to be Automatic - set up a new location with NO customisation. Every try improved the time between quality going from "Good to Best" (Amazon video definition).

But, it still seemed to happen - on all browsers, Chrome, Firefox, Brave and Safari. So it must be my MAC - where next? Well, watching Amazon Sport using a Private browser Window seemed to be the answer. What is loading and causing Safari to loose video quality? It's not the few extensions (switched off-even uninstalled) - it's not WiFi (used Ethernet connection the router) but, Private browsing fixes the problem - why? Don't know.

It's true unfortunately, I'm afraid I was too premature in reporting the workaround. I also had big hopes when I found that >Preferences >battery >Battery >uncheck "Optimise video streaming while on battery". Unfortunately, all these things seem to improved the time between video quality degrading episodes, but they don't fix the problem permanently.

Problem is, that for me, Safari is the browser I prefer and I want it to work. I'm sure that I have observed this HD degrade in Brave, FireFox and Chrome - so maybe we should be concentrating on the MAC. Why I say that is; "sometimes" when selecting a web video to play, the target tab turns black and there is no activity - if you then HotKey "option + Command + E" - you may need to have activated the Develop Menu is Safari Preferences (last item on the Advanced Tab), then as soon as the cache files are cleared the busy ring starts running and the video loads. May be a helpful clue?

yea this is nuts. even since jobs left us ,apple is screwing the pooch waaay too often . and i wish people would stop giving instructions to re-configure the entire setup or change an app or some crazy thing.

90f70e40cf
Reply all
Reply to author
Forward
0 new messages