WIFImon NORDUnet Tech Workshop 2015

WIFIMon

“How to fix WIFI take 2”Dave Wilson, James Healy, Kostas Stamos, Vasileios Kokkinos, Arne

Øslebo, Kurt Baumann, Jan Hertzberg, Brian Bach Mortensen (presenter)

Géant4 - Service Activity 3 Task 3

Problem statement

• Users complain about WIFI performance to local network admin

• Typical solutions:• Network admin checks AP logs or hardware probes (if any)• Send someone to check the installation

• Problem might be solved due to students leaving the room

• If “hard” fault –>• No Problem case solved done

• If “soft” problem à• more data is needed to determine the cause and the fix

• Furthermore, a proactive solutions was sought

Road to a fix

• Problem was raised at the Dublin City University (James Healy)

• They contacted HEAnet (Dave Wilson) thinking that they would have a good-cheap-fast fix to the problem

• Getting all 3 at the same time turned out to be problematic

• HEAnet went to Géant with the idea:• Lets crowd source the performance of the WIFI

installations

Measuring WIFI 1.

• Hardware probes are great for WIFI monitoring

• But they are typically fixed a certain locations

• Users are not!

• They measure probe performance• Users don’t care about probe performance

Measuring WIFI 2.

• Software probes can be run by user device

• They are cheap

• They are abundant (think student devices)

• They (may) represent the performance the user gets

What we need then?

• JavaScript that runs in the users browser that measures

• Bandwidth/Latency• IP Address of the user• Timestamp

What we need then (cont.)?

• DHCP/Radius logs• Mac address of user and access point• IP Address of the user• Timestamp

• This will make it possible to correlate JavaScript measurements with Access Points• Pin pointing problems to specific APs

Client Software

• Rather than implementing measurement JavaScript from bottom we took a look at open source implementations:• Nettest

(https://code.google.com/p/nettest/)• Boomerang

(https://github.com/yahoo/boomerang)• Currently using nettest

• Planning to test Boomerang

Correlation

• Currently we are using Nettest• Correlating the data from JavaScript

with Radius/DHCP • Lets see a simple check:

Basic checking

0

500

1000

1500

2000

2500

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

Download rate(KB/s)

Upload rate(KB/s)

Measuring at NT2015

• Not optimal conditions :0(• Only hotel Wi-Fi• No log files to correlate measurements

from JavaScript to AP points• Website did not implement JavaScript

on schedule page• No time to fix it before people where here

• But we did get a few measurements tru…

• 48 measurements as of last night

Bandwidth

• Clearly not all measurements are “good”

Ping

Ping

• Correlation between bad ping and bandwidth• User at edge of Wi-Fi reach?

Statistics

• Mean values• Down, Up, Ping:• 1107KB/s, 476KB/s, 47ms

• Standard deviation• 688KB/s, 194KB/s, 30ms

• Pretty big variance from the mean• So maybe not that good to use for

anomaly detection??

Statistics cont.

• So if the above is normal behavior• How do we detect issues with the Wi-Fi network?

• We probably need to get data from healthy and unhealthy scenarios• Then we can compare the stats and see if there

is big changes in mean/variance etc.• E.g. if enough pings go wrong it will eventually shift

the mean value to a higher point• And perhaps the variation will go down?

• This is of course so far PURE guesswork from my side!

Further Data Validation

• Comparing Data with HW probes is a topic that might reveal the usefulness of WIFImon (and probes)

• Will the probes and the clients see the same degradation

• Or will one “party” have an edge?• Whatever the result it might be a good

way to detect problems to correlate the varios measurements

Summary

• We got the basis for doing a more throughout validation

• We do need a lot more data to make any attempt on drawing conclusion

• If you want to participate with data collection• Do not hesitate to contact us

• We have a small agile team• Changes are made on a daily basis

• With your help we will know if worth continuing!

Thanks

• All from the WIFIMon team• Erik Kikkenborg

• Feel free to contact me or any team member with ideas, suggestions, corrections for further work

E-mail: [email protected]