Trace-Driven Analysis of ICN Caching Algorithms on Video-on-Demand Workloads

Abstract

Even though a key driver for Information-Centric Networking (ICN) has been the rise in Internet video traffic, there has been surprisingly little work on analyzing the interplay between ICN and video -which ICN caching strategies work well on video workloads and how ICN helps improve video-centric quality of experience (QoE). In this work, we bridge this disconnect with a tracedriven study using 196M video requests from over 16M users on a country-wide topology with 80K routers. We evaluate a broad space of content replacement (e.g., LRU, LFU, FIFO) and content placement (e.g., leave a copy everywhere, probabilistic) strategies over a range of cache sizes.We highlight four key findings: (1) the best placement and replacement strategies depend on the cache size and vary across improvement metrics; that said, LFU+probabilistic caching [37] is a close-to-optimal strategy overall; (2) video workloads show considerable caching-related benefits (e.g., ≥ 10% traffic reduction) only with very large cache sizes (≥ 100GB); (3) the improvement in video QoE is low (≤ 12%) if the content provider already has a substantial geographical presence; and (4) caches in the middle and the edge of the network, requests from highly populated regions and without content servers, and requests for popular content contribute most to the overall ICN-induced improvements in video QoE.

Exploiting Caching and Multicast for 5G Wireless Networks

“…To model the user demand in our evaluation, we uniformly spread the requests in the trace across the coverage regions of the 14 SBSs. We further spread these requests across files using a Zipf popularity distribution with shape parameter z [40]. This results the demand values λ ni for each SBS n and file i.…”

“…Unless otherwise specified, all files are of size 30MB and each SBS is equipped with a cache that can store up to 20% of the entire file library size. Finally, we set z = 1.2 (as in [40]) and d = 3 minutes, while our evaluation also covers a wide range of z and d values.…”

Scalable Cache Management for ISP-Operated Content Delivery Services

“…In [28], Sun et al indicate that, in their VoD dataset, the size of the catalog of watched videos amounts to 500K unique items. To satisfy the scalability constraint of our simulator while preserving a good approximation of a realistic content catalog, we consider a list of 100K unit sized video content items.…”

“…For each content, the locations are randomly selected among the 88 edge nodes and the number of requests for the item is then equally divided between these locations. We select the value of z based on the characteristics of the dataset used in [28] and set it to 1.174. The value of β is set to 2, since according to [7] such a value increases the heterogeneity of the GDI.…”

“…In the extreme case where z is in the order of 0.1 − 0.2, the hit ratio drops to 15% − 20%. However, measurement-based studies, such as the ones presented in [30], [31], suggest that the Zipf parameter z for web traffic lies in the range of 0.64 − 0.84, while in [28] the authors show that this parameter is equal to 1.174 for an online video content provider. This implies that for real world environments, where the distribution of content popularity is less uniform (especially in video delivery systems), the proposed clustering and placement schemes perform significantly well.…”