Boosting Streaming Video Delivery with WiseReplica - Transactions on Large-Scale-Data-and Knowledge-Centered Systems XX - page 46

Table 2. Default values for workload parameters.

Workload
Requests per user uniform

Experiment duration 4h

Mean requests per second 100

Requests fractions 5 % of creations, 95 % of views

Video size (follows Pareto) shape = 3, between 13 MB and 1.6 GB

Video popularity (Zipf-Mandelbrot) shape = 0.8, cutoff = number of videos

Videos' creation (Poisson) λ = creations per second

Popularity growth from YouTube traces 21827 distinct patterns
YouTube encoding settings (bitrates)

5 Mbps, 15 Mbps, 30 Mbps, 50 Mbps

parameters. Finally, videos are always divided and distributed in chunks or seg-
ments of fixed size, 2 MB.
In terms of SLA definition, we assume that content and content delivery
providers are committed to improving the Internet video availability for cus-
tomers in a content-oriented approach. In our case, a good peer-assisted VoD
system must ensure videos availability by avoiding rebuffering. Therefore, we
consider a global, simple SLA contract drawn up to provide a minimal average
bitrate according to each Internet video encoding setting. A SLA violation hap-
pens whenever the system fails to enforce the minimal average bitrate for any
viewer session.

5.2 Evaluation Scenario

Our evaluation scenario (Fig. 3 ) includes 4002 nodes, arranged across two stor-
age domains. There are one coordinator and 2000 peers per storage domain.
Storage and network capacities differ according to the device role. Coordinators
have 20 TB of storage capacity and full-duplex access link of 4 Gbps. Peers con-
tribute 200 GB each, equipped with 100 Mbps full-duplex links. Note that the
two coordinators contribute with a small fraction of aggregate edge resources,
i.e. 5 % of the storage capacity and only 2 % of the total network capacity. This
draws our attention to the performance of replication schemes towards peers
resource allocation. We assume only 1 % peers' storage is available for caching
additional replicas, namely 2 GB.
We implemented and evaluate this work using simulation. To this end, we
developed a simulation tool on top of PeerSim [ 24 ] to implement storage domains
in edge network and bandwidth scheduling.Our design focus on network's re-
source allocation accuracy for simulating bitrate enforcement and concurrent
videos views properly. Design and implementations details of our tools to sim-
ulate network resource scheduling are available in our previous work [ 28 ]. We
have performed our simulations using servers equipped with Intel Xeon E5450
3.00 GHz, and a RAM of 4 GB.