Add documentation from our design doc to README
This makes the project more self-contained, and will help others using
it to reproduce our results if needed.
diff --git a/wifitables/README b/wifitables/README
deleted file mode 100644
@@ -1,5 +0,0 @@
-WifiTables constructs tables of WiFi performance vs distance.
- ./sample at each location you want to gather measurements for.
- ./report at the end of a series of runs to collate the data into a table.
diff --git a/wifitables/README.md b/wifitables/README.md
new file mode 100644
@@ -0,0 +1,214 @@
+# wifitables - construct tables of Wifi performance vs distance
+Developing access points requires knowing what wireless chips, and
+access points using them, do under real-world conditions of low signal strength
+and high congestion. Existing utilities are good at
+gathering the information needed to understand wireless network performance:
+ * [airport](http://kb.mit.edu/confluence/pages/viewpage.action?pageId=4272001)
+ on Mac, and
+ [iw](https://wireless.wiki.kernel.org/en/users/documentation/iw) on Linux,
+ to gather information on signal strength and noise
+ * [tcpdump](http://www.tcpdump.org) for information on wireless connection
+ * [iperf](http://sourceforge.net/projects/iperf/) for TCP and UDP throughput
+but using them to gather information over dozens of test runs would be tedious
+and error-prone. We wrote `wifitables` to automate this for us. Once it is
+set up, running tests requires only a few commands:
+> `./sample` at each location you want to gather measurements for
+> `./report` at the end of a series of runs to collate the data into a table
+wifitables assumes you're using a similar test environment to what we use,
+ * a MacOS or Linux laptop as a test client
+ * a Linux workstation as a test server
+# Design Philosophy
+wifitables is specifically designed for experiments with signal strength
+and network congestion. These are the real-world variables that are most likely
+to affect end users' experiences, and to be overlooked by existing test
+To control signal strength, we move one of the devices in our test network
+between tests. Increasing the distance between device and AP decreases signal
+strength. For our tests, we marked about a dozen test points through the office
+floor, starting close to my desk and extending to the far end of the floor.
+A busy office provides a natural experiment for congestion: both 2.4GHz and 5GHz
+(non-DFS) bands are heavily congested during the day, and lightly congested at
+night, so we repeat data collection at the same set of distances during day and
+# Setting up wifitables
+You will need to install some software on your workstation, and on each
+computer you’re using as a test client. You don’t need to install any software
+on GFiber equipment, since it is already included in the system image.
+## On your workstation:
+ 1. Install iperf
+ $ sudo apt-get install iperf
+## On each test client you want to use:
+ 1. Install iperf. On a Mac with Homebrew installed:
+ $ brew install iperf
+ Commands for other platforms will vary.
+ 2. Copy wifitables to your test client. You can either clone the git
+ repository, or download an archive from Gitiles and extract it. No further
+ installation steps are required.
+ 3. On a stock MacBook Pro, you’ll need to [symlink `airport` into the system
+ path](http://apple.stackexchange.com/a/130924) for the `sample` script to
+# The essential test network
+Our access point tests take place across a local area network. This network
+includes a minimum of three hosts:
+ * A Fiber Network+ Box (GFRG2x0) to act as a router for the network.
+ * A host running iperf2 in TCP server mode, and running iperf2 in UDP server
+ mode, often a workstation in the lab.
+ * A laptop running test client code. This is most thoroughly tested on a
+ MacBook Pro, but Thinkpads running Ubuntu have worked as well.
+The WAN link of the Network+ Box is connected to a wired network with public
+Internet access; this is needed in order to manage it with the Network Tab.
+Within Google, a lab network should be used when available.
+The LAN links of the Network+ Box are connected to different hosts depending on
+what the objectives of the test are. These configurations are documented for
+each test network.
+# Networks for Access Point wireless testing
+These networks can be tested fully with the `sample` utility in wifitables.
+## Network+ Box
+![Network block diagram](docs/NetworkBoxTesting.png)
+Testing Network+ Box wireless performance requires only the elements of the
+essential test network. To physically build this network:
+ 1. Connect your Network+ Box to the Internet.
+ 2. Connect your workstation to your Network+ Box on a wired connection.
+ 3. Use the Network Tab in the Fiber Portal to set up a wireless network.
+ * It’s strongly recommended to enable separate 2.4 GHz and 5 GHz SSIDs.
+ 4. Join the wireless network from your laptop.
+ 5. Make sure your laptop has Internet access and can ping your workstation at
+ its IP address _on the test network._
+## Any other Access Point
+![Netwok block diagram](docs/AccessPointTesting.png)
+Testing wireless using another Access Point requires a bit more effort to get
+set up, and the effort required varies by manufacturer.
+ 1. Connect the Network+ Box to the Internet, and the workstation to the
+ network box, if they are not already connected. Refer to steps 1 and 2 in
+ [Testing Network+ Box wireless](#network_box) if you haven’t done this
+ 2. In the Fiber Portal, turn *off* wireless networking on the Network+ Box.
+ Two access points in close proximity (roughly, two meters or less) will
+ interfere with each other, and you won’t get accurate test results.
+ 3. Connect the AP under test to the Network+ Box using a wired connection.
+ Power it on if it has a power switch.
+ 4. Perform initial configuration of the Access Point so it:
+ * serves a 5 GHz network, and
+ * is operating as a pure access point, not a router.
+Initial configuration is vendor-specific; for some, it will be possible to do
+over a computer wired to the Network+ Box. For others, it will be necessary to
+join their wireless network in order to do this.
+If the AP under test offers a firmware update during the setup process, accept
+Once you have completed initial configuration:
+ 1. Join the AP's wireless network from your test laptop.
+ 2. Make sure your laptop has Internet access and can ping your workstation.
+# Running a test series
+Once you’ve installed the required software, and set up your test network,
+you’re ready to start testing.
+_Replace_ 192.168.1.2 _with the IP address of your own workstation on the test
+network when working through these procedures._
+ 1. Start the `iperf` servers on your workstation. I recommend running them
+ under `screen` or `tmux` so they can stay running reliably:
+ $ iperf -sB 192.168.1.2 -w 1M & $ iperf -usB 192.168.1.2 &
+ 2. Determine the IP address of your workstation as seen by the Network Box.
+ 3. On your test client, choose a directory that the data gathered by the test
+ script will live in. Create it if it doesn’t already exist. I use one that
+ syncs to a folder in Google Drive so the data gets shared with my team
+ $ mkdir -p path/to/datadir
+ 4. Choose a name for your test series. I usually choose something descriptive
+ of the test performed, like `gfrg200-day` for a test of a Network Box (model
+ number GFRG200) performed during the daytime.
+ 5. Place your test client one step away from the Access Point that you’re
+ 6. On your test client, change into the wifitables directory; if you
+ extracted the tarball in the current directory, the command will look like:
+ $ cd test-master/wifitables
+ 7. Execute the sample command, specifying the journal (-j) as the name of
+ your test series, the destination (-d) as your workstation, and the number of
+ steps (-s) as 1 since you’re one step away.
+ $ ./sample -j path/to/datadir/gfrg200-day -d 192.168.1.2 -s 1
+ 8. Enter your password for `sudo` if prompted; it’s required on
+ some systems to capture wireless traffic.
+ 9. Wait for the test to complete.
+ 10. If this is your first series in a new environment:
+ 1. step back a few steps - usually between 5 and 10 will work.
+ 2. find a convenient place to set or hold your test client.
+ 3. mark it with some tape and the number of steps you took.
+ 4. set your test client down if you were going to do that.
+ 11. If it’s not, step to the next place you have marked.
+ 12. Edit the sample command to reflect the new number of steps you’re at and
+ run it:
+ $ ./sample -j path/to/datadir/gfrg200-day -d 192.168.1.2 -s 6
+ 13. Go back to step 9. Repeat until your MCS drops to 0 (you no longer have a
+ signal) or you run out of room to step backwards.
+# Reporting on a test series
+Once you have gathered data for a series of tests and want to report on it, you
+can run the `report` utility from wifitables. In normal use, this
+takes a single positional argument: the series you want to report on. It will
+write the report, in tab-delimited format (.tsv), to standard output. You can
+redirect this to a file if you want.
+To report on the series collected earlier, you’d run
+ $ ./report path/to/datadir/gfrg200-day
+Tab-delimited text can be imported automatically by Google Sheets if you copy
+and paste it into a newly created worksheet; this is the workflow that we're
+using with this script at Google.
diff --git a/wifitables/docs/AccessPointTesting.png b/wifitables/docs/AccessPointTesting.png
new file mode 100644
Binary files differ
diff --git a/wifitables/docs/NetworkBoxTesting.png b/wifitables/docs/NetworkBoxTesting.png
new file mode 100644
Binary files differ