UW-CSE logo University of Washington Department of Computer Science & Engineering
 Wireless (Wi-Fi) networks
 CSE Home About Us    Search    Contact Info 

  Ed Lazowska
  Ratul Mahajan
  Charles Reis
  Maya Rodrig
  David Wetherall
  John Zahorjan
    We are trying to understand how well 802.11 networks work in practice and how they can be improved.


  • Download Wit, our tool to analyze wireless MAC. This is an almost pre-alpha release. We'll appreciate your feedback.

    The README file that comes with the code.


  • Measurement-Based Models of Delivery and Interference in Static Wireless Networks [ps | pdf]
    Charles Reis, Maya Rodrig, Ratul Mahajan, David Wetherall, and John Zahorjan
    ACM SIGCOMM Sept, 2006
  • Analyzing the MAC-level Behavior of Wireless Networks [ps | pdf]
    Ratul Mahajan, Maya Rodrig, David Wetherall, and John Zahorjan
    ACM SIGCOMM Sept, 2006
  • Measurement-based Characterization of 802.11 in a Hotspot Setting [ps | pdf]
    Maya Rodrig, Charles Reis, Ratul Mahajan, David Wetherall, and John Zahorjan
    ACM SIGCOMM E-WIND Workshop Aug, 2005


  • Collection topology
    • image provides details on placement but is not to scale; see the E-WIND paper for correct scale
    • AP channels:
      • channel 1: sigcomm-nat,sigcomm-public-3,sigcomm-public-1
      • channel 8: sigcomm-nat-foyer
      • channel 11: sigcomm-public-2
    • traces from monte were corrupted in the collection process
  • Data   (tcpdump of prism headers, gzipped)
    • we used the noise field of the prism frame to encode error status of frames, so be sure to filter down to only those frames with noise values of zero. others are errored packets (corrupt or not fully received).
    • monitor channels (primarily): ath0=1, ath1=8, ath2=11. exception is that the APs switched from channel 8 to channel 6 on Wednesday morning (9/1/2004); kalahari switched ath1 to channel 6 Wednesday at 12:24 pm, and sahara, chihuahuan, and sonoran switched ath1 to channel 6 Friday at 8:30 am.
    • MAC addresses and IP addresses have been anonymized using a tool from CRAWDAD project
      • anonymization is consistent across trace files (IP-x maps to IP-y in all of them)
    • the clocks on different machines were not synchronized; chihuahuan's clock is the most accurate
    • contents above the transport layer have been removed
    • note: three of the ath2 traces (chihuahuan, mojave, and sonoran) were very large. the split command was used to break them up into 1G segments to allow sharing them online. you will need to cat the segments of each before unzipping.

CSE logo Last updated: Aug, 2005
[comments to Ratul Mahajan]