Migrating Backoff to the Frequency Domain
Total Page:16
File Type:pdf, Size:1020Kb
No Time to Countdown: Migrating Backoff to the Frequency Domain Souvik Sen Romit Roy Choudhury Srihari Nelakuditi Duke University Duke University University of South Carolina Durham, NC, USA Durham, NC, USA Columbia, SC, USA [email protected] [email protected] [email protected] ABSTRACT 1. INTRODUCTION Conventional WiFi networks perform channel contention in Access control strategies are designed to arbitrate how mul- time domain. This is known to be wasteful because the chan- tiple entities access a shared resource. Several distributed nel is forced to remain idle while all contending nodes are protocols embrace randomization to achieve arbitration. In backing off for multiple time slots. This paper proposes to WiFi networks, for example, each participating node picks a break away from convention and recreate the backing off op- random number from a specified range and begins counting eration in the frequency domain. Our basic idea leverages the down. The node that finishes first, say N1, wins channel con- observation that OFDM subcarriers can be treated as integer tention and begins transmission. The other nodes freeze their numbers. Thus, instead of picking a random backoff duration countdown temporarily, and revive it only after N1’s trans- in time, a contending node can signal on a randomly cho- mission is complete. Since every node counts down at the sen subcarrier. By employing a second antenna to listen to same pace, this scheme produces an implicit ordering among all the subcarriers, each node can determine whether its cho- nodes. Put differently, the node that picks the smallest ran- sen integer (or subcarrier) is the smallest among all others. dom number transmits first, the one that picks the second- In fact, each node can even determine the rank of its chosen smallest number transmits second, and so on. The overall subcarrier, enabling the feasibility of scheduled transmissions operation is often termed as “backoff”. after every round of contention. We develop these ideas into a Back2F protocol that migrates WiFi backoff to the frequency While backoff arbitrates channel contention, it incurs a per- domain. Experiments on a prototype of 10 USRPs confirm fea- formance cost. Specifically, when multiple nodes are simulta- sibility, along with consistent throughput gains over 802.11. neously backing off, the channel must remain idle, naturally Trace based simulations affirm scalability to larger, real-world leading to under-utilization. Moreover, network congestion network topologies. prompts an exponential increase in the backoff range, intro- ducing the possibility of greater channel wastage. Authors in [16] show more than 30% reduction in throughput due to backing off; [13] shows the severity at higher data rates. This paper attempts to address this problem by migrating the backoff operation to the frequency domain. Categories and Subject Descriptors Our main idea is simple. When a node N has a packet to C.2.1 [Network Architecture and Design]: Wireless commu- 1 transmit, it picks a random value, r , from a specified range nication 1 [0;F ]. Once the channel becomes idle, N1 begins the back- off operation. However, instead of counting down from r1 th 1 General Terms to 0, N1 transmits a symbol on the r1 subcarrier . We as- Design, Experimentation, Performance sume that each node has two antennas; thus, while one an- tenna transmits, the other antenna listens to determine which of the subcarriers are active. Assuming N2 is also contending th Keywords for the channel, and say has transmitted on the r2 subcarrier, Wireless, Contention Resolution, Cross-Layer, Backoff N1 observes activity on both subcarriers r1 and r2. Assuming r1 < r2, N1 immediately infers that it has won channel con- tention, and begins transmission. N2 learns that it has lost, and defers its own transmission until N1 has finished. We call this approach Back2F, as an acronym for migrating backoff to the frequency domain. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are The advantages of Back2F are two fold. First, one round of not made or distributed for profit or commercial advantage and that copies frequency domain backoff should ideally last for few OFDM bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific symbols, substantially less than the average backoff in proto- permission and/or a fee. 1 MobiCom’11, September 19–23, 2011, Las Vegas, Nevada, USA. Subcarriers are narrowband OFDM channels used by Copyright 2011 ACM 978-1-4503-0492-4/11/09 ...$10.00. 802.11. cols like 802.11. Second, Back2F creates a logical ordering We make three observations that are not necessarily new. (1) among contending nodes, and each node learns its own rank Fundamentally, backing off is not a time domain operation. in this order. This ranking among nodes creates the possi- Its implementation is in the time domain, forcing the chan- bility of batched transmissions, eliminating the need for per- nel to be idle before each packet transmission. (2) The du- packet backoff. Since 802.11 currently backs off before every ration of each backoff slot is fixed, so the channel wastage packet, Back2F helps improve the channel usage and network grows with smaller packets and higher bitrates. A packet’s throughput. airtime is shorter at higher rates, and hence, the fraction of channel-time occupied by idle slots is larger. (3) Finally, Of course, extracting these gains entail a number of research although channel utilization may improve with few nodes challenges: (1) Active subcarriers need to be detected accu- (backing off in parallel), just a few more nodes can cause col- rately in face of loose time synchronization among transmit- lisions (802.11 experiences 18% collisions with 3 nodes and ters, energy leakage between narrow-band subcarriers, and CW =16). A collision forces nodes to exponentially increase channel fading. (2) Collision among nodes – which hap- their backoff, pushing the system back to under-utilization. pens when multiple nodes choose the same subcarrier – needs Fig. 1 shows the channel under-utilization due to 802.11’s to be mitigated successfully. (3) Finally, since nodes are lo- backoff, under varying bitrates and network densities. Au- cated in different contention neighborhoods, the node rank- thors in [16, 29] corroborate these findings with extensive ings do not obey any global order. Back2F needs to cope analysis and measurements, emphasizing the need to improve with relative ranking among nodes, while maintaining spatial wireless contention resolution. reuse and fairness comparable to 802.11. We address these challenges through activation of diverse subcarriers, multiple rounds of contention, and virtual countdown. We consolidate 0.45 these ideas into a protocol, and implement a prototype on the 0.4 USRP/ GNURadio platform. Experimental results show 95% accuracy in subcarrier detection, less than 2% probability of 0.35 collision, and throughput gains of more than 35% at high bit- 0.3 rates. Trace driven simulations (with topologies and channel 0.25 conditions drawn from our university network) confirms sta- bility and scalability of Back2F to real-world scenarios. 0.2 2 APs 512 bytes 0.15 Our contributions in this paper may be summarized as: 6 APs 512 bytes 0.1 10 APs 512 bytes • We identify an opportunity to migrate protocol operations 2 APs 1024 bytes from the time to the frequency domain. Although we in- 0.05 6 APs 1024 bytes 10 APs 1024 bytes stantiate our ideas through a WiFi based MAC, they may 0 be generalized to other arbitration strategies. Channel waste due to backoff(%) 10 15 20 25 30 35 40 45 50 55 • We design an OFDM based system where random backoff is Bitrate in Mbps realized by selectively transmitting on a subcarrier. A log- ical order among senders is enforced in a decentralized Figure 1: Overhead of 802.11 backoff. Larger fraction of manner, for improved channel usage. channel wasted with smaller packets at higher bitrates. • We address the challenges behind such a scheme, and pro- totype it on the USRP/GNURadio platform. Promising re- Orthogonal Frequency Division Multiplexing: OFDM can sults, in terms of throughput, fairness, and scalability, be abstracted as a PHY scheme that divides the wireless spec- give us confidence to build a larger system. trum into multiple narrow band channels, called subcarriers. The subcarriers carry modulated data streams in parallel, but 2. 802.11 AND OFDM at a lower rate per-subcarrier. The benefit of OFDM emerges This section highlights the limitations of 802.11’s backoff (in from its ability to cope with channel adversities, including time domain), and presents a simple abstraction of OFDM (to narrowband interference and frequency-selective fading due better explain the shift to frequency domain). to multipath. The 802.11a/g implementation of OFDM has 52 subcarriers, of which 48 are used for data transmission, 802.11 Channel Access: WiFi prescribes each transmitter to and 4 for equalization. A transmitter stripes bits across all backoff for a random number of slots, chosen from the range subcarriers, however, it is possible to transmit/receive only [0;CW − 1], where CW is the contention window. Each time on a subset of them. slot corresponds to 9µs. The node counts down only if the channel is idle – if the node senses a busy channel, the count- As we will see later, a Back2F node picks a random number, down is frozen, and revived only after the channel is idle. say 11, and transmits a short signal only on the 11th subcar- Whichever node completes the countdown first begins trans- rier. The node’s second antenna detects a strong signal on the mission.