Peer-Based Collaborative Caching and Prefetching in Mobile Broadcast - Advanced Operating Systems and Kernel Applications - page 253

will not receive any announcement on prefetching
of D 1 so it will decide to prefetch D 1 .
From this example, we see that in cooperative
prefetching a mobile peer may rely on its neigh-
bors for a data object, because the mobile peer is
monitoring the broadcast channel continuously
and once the data object is not available at its
neighbors the mobile peer may prefetch it when
it appears again, so that costly cache misses are
avoided.
Nevertheless, CPIX and ACP share one im-
portant point in common: they keep the mobile
peers autonomous even though each peer takes
the data availability at its neighbors' into account
when making caching and prefetching decisions.
This ensures that small change in neighborhood
will not affect CPIX and ACP. This is achieved
by not relying on specific neighbors when making
caching and prefetching decisions. Rather, a peer
makes a decision based on its whole impression
of its neighbors. In CPIX, a mobile peer makes
caching decision totally based on local statistics,
and it tolerates the changes of neighbors. In ACP,
a mobile peer makes prefetching decision based
on the number of announcements it receives. If a
peer M a makes a prefetching announcement for a
data object but moves away soon, it will not have
a significant effect on its neighbors because they
do not rely on M a specifically.

of GOP and SOP because they are designed for
mobile peers that form stable groups while we are
interested in scenarios where each mobile peer
follows its own trajectory.

Simulation model

The simulated mobile environment is an X*Y
(m 2 ) area where there are a broadcast server
and n mobile peers. The server broadcasts data
objects to the mobile peers through a wireless
channel. The bandwidth of the broadcast chan-
nel is bb Mbps. All mobile peers can receive
data objects that the server broadcasts. A mobile
peer can communicate with another peer if they
are in each other's communication range. Their
transmission range is TransRange meters. The
bandwidth of the short-range communication is
sb Mbps. At the beginning, the mobile peers are
randomly scattered in the area. The mobile peers
then move in the area following a variant of the
“random waypoint” mobility model.
In the model (and experimental results), we
use a time unit called broadcast unit . A broadcast
unit is the time the server takes to broadcast one
data object.

Broadcast Server

We adopt broadcast disk s (Acharya & Alonso,
1995) to model the server's non-uniform broad-
cast. The server has m broadcast disks, and they
are Disk i where 1≤i≤m . Disk i stores DiskSizei i data
objects and spins at a speed of DiskSpeedi i . All
data objects are read-only and of the same size
which is DataSize KB. As in (Acharya & Alonso,
1995), we use a parameter Δ to capture the rela-
tive speeds of the disks: DiskSpeedi i = (( m - i )*Δ)
+1). For example, if the server has 3 broadcast
disks and Δ is 2, then the rotation speed of the
disks will be 5, 3 and 1. Δ is used to model the
nonuniformity of the broadcast. When Δ is 0,
the broadcast is uniform. The bigger the Δ is, the
more non-uniform the broadcast is.

eValuation of
collaboratiVe caching
and prefetching SchemeS

To study the performance of CPIX and ACP, we
conducted detailed simulation experiments. We
report representative results here. The reader may
find more details of the experiments and results
in (Wu, 2005; Wu, 2006).
In the experiments, we compare CPIX, ACP
with DGCoca (Chow, 2005), a cooperative cach-
ing scheme designed for push-based broadcast
environment. We did not study the performance