Boosting Streaming Video Delivery with WiseReplica - Transactions on Large-Scale-Data-and Knowledge-Centered Systems XX - page 47

Fig. 3. Evaluation scenario.

5.3 Comparable Replication Schemes

We compare WiseReplica with two other schemes.

Non-collaborative Caching. Adaptive replication schemes based on non-
collaborative caching, such as those that uses Least Recent Used (LRU) algo-
rithm, are easy to implement and deploy. A new replica is created in a peer
whenever a user requests to view a video. LRU replacement is enforced regard-
ing the static percentage of the local storage capacity for caching of 1 %.

Oracle-LikeCollaborativeCaching. This is an idealized benchmark case. Here,
we assume a peer-assisted VoD system deployed in a network that runs a deadline-
aware transport protocol, similar to Wilson et al. [ 34 ] work. Based on our previous
work with AREN [ 29 ], an adaptive replication scheme for edge networks, we imple-
mented a benchmark replication scheme that relies on bandwidth reservation and
collaborative caching to provide an adaptive number of replicas for videos. We
replicate videos according to aggregate network usage by enforcing a low and high
thresholds. This makes the video replication a function of bandwidth reservation,
and ensures that network and storage provision follows video demand properly, as
depicted in Fig. 4 . Per video, we consider two percentage thresholds for aggregate

Fig. 4. Oracle-like bandwidth management for a video, illustrating aggregate band-
width for N replicas and b available bandwidth, bandwidth reservation (bandwidth

usage) and thresholds ( P min and P max ).