Resource Allocation Techniques to Improve the Performance of Wireless Networks
Wireless networks are increasingly becoming important in enabling convenient Internet access
for users. Many users have WiFi networks in their homes and have access to cellular networks
(e.g., 3G, 4G) when they are on the go. Unfortunately, wireless networks do not offer performance
guarantees. The download and upload rates achieved with these networks fluctuate and may often
degrade to the point where they become practically unusable. One of themajor factors that causes
this performance degradation is interference, which can be defined as the unwanted noise generated
by other devices in the wireless spectrum.
The root cause of interference lies within the fact that the open nature of wireless communications
enables multiple stations to access the spectrum simultaneously. To eliminate interference,
wireless communication systems offer multiple orthogonal channels. If multiple stations access
the spectrum using these orthogonal channels, interference is very likely to be avoided (assuming
that the signal "leak" between orthogonal channels is negligible), as opposed to the case where
stations use the same channel. Therefore, one needs to determine the specific access patterns (i.e.,
which channel is used by which transmitter at what point in time) to systematically coordinate the
transmissions of multiple wireless stations.
While a large body of previous work exists to coordinate the use of orthogonal channels between
IEEE 802.11 a/b/g stations, we show that they are not able to address the interference problem for
IEEE 802.11n devices due to the use of wide channels in 802.11n systems. Thus, we propose
a WLAN management system - called ACORN - that not only allocates orthogonal channels to
802.11n access points but also makes intelligent user association decisions to significantly improve
aggregate WLAN throughput. Later, we study the interference in cellular networks, in particular
OFDMA femtocell systems. We experimentally characterize the interference between OFDMA
femtocells and propose design guidelines that help in building resource management solutions to
alleviate interference. We observe that resource management solutions for OFDMA femtocells need
to implement unique functionalities due to the fundamental differences between OFDMA access
technology and 802.11 WiFi systems. Leveraging the guidelines revealed by our study, we later
propose FERMI - a resource management solution that mitigates interference between OFDMA
femtocells and at the same time leverages spatial reuse opportunities. We implement FERMI on
a WiMAX femtocell testbed and show that it provides throughput benefits as much as 7x over a
baseline scheme and it is closer to the optimal solution than its simpler theoretical alternatives.
In addition to the multi-cell interference setting addressed by our above-mentioned contributions,
we study beamforming in a single cell scenario. Beamforming is a technique that allows
focusing the energy of transmissions in a particular direction to improve overall signal quality.
Since OFDMA schedules multiple users in the same frame (and this is in contrast toWiFi), designing
data scheduling solutions and at the same time benefiting from beamforming is challenging. We
propose iBUS - a system that addresses uplink data scheduling with beamforming. We show that
iBUS is provably efficient in striking the right tradeoffs inherent in this setting. Evaluations from
both a prototype system implementation and simulations yield that iBUS improves the throughput
by more than 40% compared to an omni-directional scheduling scheme.
Finally, we study distributed computing on smartphones. We identify that there are a large
number of phones that are left idle (often when being charged overnight) for long periods of time.
We envision that an enterprise could use such idle smartphones (e.g., phones of its employees) for enabling parallel computing on them. We believe that smartphones offer precious computing
resources that have not previously been tapped into from a distributed computing perspective. In
addition to the computing capacity being offered, our cost analysis and projections show that smartphones
are also an energy-efficient alternative to using dedicated servers for computing purposes.
However, enabling distributed computing on a large number of phones is complicated due to the
unique set of challenges that smartphones present, such as heterogeneity in CPU speed and variability
in network bandwidth. We design and implement CWC - a distributed computing infrastructure
using smartphones. Our extensive evaluations demonstrate that CWC can schedule tasks for 1.6x
faster completion that alternative mechanisms.