Boosting Streaming Video Delivery with WiseReplica - Transactions on Large-Scale-Data-and Knowledge-Centered Systems XX - page 54

curves of WiseReplica and Oracle-like are quite similar indeed. It confirms that
our predictions are accurate, and that a simple replication policy works properly.
Reducing the number of replicas implies that the systems requires less storage
for replication. Figure 11 shows storage usage for replicas by replication scheme.
Although WiseReplica utilizes more storage than Oracle-like, its usage remains
two orders of magnitude smaller than a non-collaborative caching. The maximum
storage usage for Oracle-like, WiseReplica, and a non-collaborative caching were
34, 85, and 7921 GB respectively. WiseReplica creates more replicas than Oracle-
like because it does not rely on bandwidth reservation to prevent violations. Yet,
WiseReplica maintains replicas eciently, keeping storage usage very low, and
making cache replacement policies unnecessary. This suggests that the LRU
policy, which would eventually be enforced in peers cache, has no impact on the
WiseReplica performance.

Enhancing Bitrate Provision for Meeting Consumers' Expectation.
WiseReplica performance is also quite similar to Oracle-like regarding preventing
violations. Each point of the Fig. 12 represents the number of SLA violations for
intervals of five minutes. Overall, caching caused 7049 violations affecting 86 %
of all viewers, WiseReplica had just 106 violations, and Oracle-like, evidently,
none. Compared to caching, WiseReplica prevents nearly 99 % of violations.
It copes with violations by (i) creating new replicas for hot videos only, and
(ii) adapting the number of replicas according to the rank position. Vertical
lines in Fig. 12 represent the first access to the 10 videos with the worst content
provision through caching. They account for 80.62 % of all caching violations.
The appearance of these videos puts the system under heavy load, which makes
caching fail to prevent violations.
Figure 13 depicts the average bitrate for viewers of the 10 videos with the
worst content provision using caching. When caching was under heavy load,

Fig. 10. The maximum number of
replicas

Fig. 11. Storage usage for replication.

for the 1 % most popular

videos.