Wireless virtualization 2007.11.14 Soyoung Park sypark@mmlab.snu.ac.kr
Contents Wireless virtualization Virtualization technique Simple implementation Wireless virtualization on commodity 802.11 hardware Gregory smith, Anmol Chaturvei, Arunesh Mishra, Suman Banerjee, University of wisconsin, wintech 07 Conclusions
Wireless virtualization Goal To enable multiple experiments to share a common wireless network concurrently or sequentially Definitions Slicing Process of allocating a coherent subset (a slice) of physical resources to a specific experiment Virtualization Allow a single machine to implement multiple instances of a required logical resource within the same or different slices
Key issues Isolation Wireless virtualization Ensures that the resource usage of one experiment has little impact on others Coherence Transmitter : active Corresponding receivers, sources of interference : simultaneously active on appropriate channels of operation Uniqueness of nodes A particular collection of nodes either due to their unique properties or to recreate an earlier configuration
Virtualization techniques Frequency division multiple access (FDMA) Time division multiple access (TDMA) Combined FDMA and TDMA Frequency hopping (FH) Code division multiple access (CDMA) Space division multiple access (SDMA)
Frequency Division Multiple Access (FDMA) Ensures that different experiments are assigned non-interfering channels Channel switching not instantaneous If multiple cards are used, channel switching time can be avoided
Time Division Multiple Access (TDMA) Each experiment is assigned time slots Needs synchronization Number of virtual nodes per physical node increases each experiment waits longer
Combined FDMA and TDMA Each experiment is assigned Frequency Partition and Time slot A given user will always use the same frequency partition + time-slot combination Switching is not instantaneous channel switching time + context switching time
Frequency Hopping (FH) Each experiment is assigned sequence of (Frequency partition, Time slot) Similar to FDMA+TDMA Difference between the two The same user will be using different frequency partition + timeslot combination
Code Division Multiple Access (CDMA) Each experiment is assigned different orthogonal codes The choice of codes is critical E.g. interferences between experiments
Space division multiple access Each experiment is assigned space Simplest approach The size of the region depends on transmission power, channel characteristics, etc. the exact choice of transmit power is important
Virtual AP Simple implementation Logical entity that exists within a physical Access Point (AP) A single provider to offer multiple services Multiple providers to share the same physical infrastructure Access Point Access Point Virtual Access Point 1 Virtual Access Point 2 Essid:1 Ch. y Essid:2 Ch. x Ch. y Essid:1 Essid:2 Exp. 1 Exp. 2 Exp. 1 Exp. 2
Simple implementation Frequency division Two (multiple) concurrent experiments by using two (multiple) interfaces Exp. 1 Exp. 1 Exp. 2 Exp. 2
wireless virtualization on commodity 802.11 hardware Gregory smith, Anmol Chaturvei, Arunesh Mishra, Suman Banerjee, University of Wisconsin, wintech 07
wireless virtualization on commodity 802.11 hardware Work goal Design and implement a system that virtualizes a wireless network using a large-scale 802.11 testbed Focus on TDM On top of a 802.11 based wireless grid (ORBIT)
TDM context switch TDM : needs time-synchronization Process : when a context switch is triggered Stop current experiment s virtual nodes Short delay Save current wireless configuration Restore next virtual node s configuration If all physical nodes ready, start virtual node
Virtualization platform Platform : UML (user mode linux) Full virtualization platform Hosted virtualization platform Wireless driver : MadWifi Doesn t run virtual node inside device driver Virtual node use tunneling request for accessing, configuring the wireless device (virtual node host OS, ioctl)
Testbed Porting the system to the ORBIT testbed ORBIT Open access research testbed for next-generation wireless networks Large two-dimensional grid of 400 802.11 radio nodes WINLAB, Rutgers University Network testbed designed to Achieve reproducibility of experimentation Supporting evaluation of protocols and applications in real-world settings
Design - Software infrastructure Node handler Defines a scripting language for configuring, controlling, monitoring node behavior Translates script into commands and multicast them Node agent
Overseer Design - Software infrastructure Task : run experiments in a RR Master/client model Centralized policy Master overseer Embodies the virtualization policy Schedules experiments Monitors physical and virtual nodes Node overseer Receives and executes orders from the master Configuring, controlling, context switch
Design - Synchronization Synchronization Synchronize the execution of commands Ordered by deadline Synchronize with execution time High-resolution timers Synchronize between nodes NTP servers : poor quality
Design Others Network configuration Experiments will need to operate in reserved address spaces Wireless configuration Node overseer doesn t deal with the card s state in full generality saving, restoring are not perfect
Performance
Performance Context switch overhead Scalability Tested up to 5 concurrent virtual nodes Primary limiting factor computing power of the physical node Node clock synchronization, system integrity
Future work TDM on a large-scale testbed Dual interface Clock synchronization improvements Wireless configuration management
Wireless virtualization Conclusions What : a way that many experiments can run on same wireless network Why : cost reduction, easier to experiment How : FDMA, TDMA, Frequency hopping, etc. Example : Virtual AP, Frequency division Implementation on 802.11 based on TDM approach Design, performance Two major challenges Node synchronization Device state