We propose a complete framework targeting operator-driven content caching that can be equally applied to both ISP-operated Content Delivery Networks (CDNs) and future Information-Centric Networks (ICNs). In contrast to previous proposals in this area, our solution leverages operators' control on cache placement and content routing, managing to considerably reduce network operating costs by minimizing the amount of transit traffic and balancing load among available network resources. In addition, our solution provides two key advantages over previous proposals. First, it allows for a simple computation of the optimal cache placement. Second, it provides knobs for operators to fine-tune performance. We validate our design through both analytical modeling and trace-driven simulations and show that our proposed solution achieves on average twice as many cache hits in comparison to previously proposed techniques, without increasing delivery latency. In addition, we show that the proposed framework achieves 19-33% better load balancing across links and caching nodes, being also robust to traffic spikes.

Sharding is a method for allocating data items to nodes of a distributed caching or storage system based on the result of a hash function computed on the item's identifier. It is ubiquitously used in key-value stores, CDNs and many other applications. Despite considerable work that has focused on the design and implementation of such systems, there is limited understanding of their performance in realistic operational conditions from a theoretical standpoint. In this paper we fill this gap by providing a thorough modeling of sharded caching systems, focusing particularly on load balancing and caching performance aspects. Our analysis provides important insights that can be applied to optimize the design and configuration of sharded caching systems.

Sharding is a method for allocating data items to nodes of a distributed caching or storage system based on the result of a hash function computed on the item identifier. It is ubiquitously used in key-value stores, CDNs and many other applications. Despite considerable work has focused on the design and the implementation of such systems, there is limited understanding of their performance in realistic operational conditions from a theoretical standpoint. In this paper we fill this gap by providing a thorough modeling of sharded caching systems, focusing particularly on load balancing and caching performance aspects. Our analysis provides important insights that can be applied to optimize the design and configuration of sharded caching systems.

We question the widely adopted view of in-network caches acting as temporary storage for the most popular content in Information-Centric Networks (ICN). Instead, we propose that in-network storage is used as a place of temporary custody for incoming content in a store and forward manner. Given this functionality of in-network storage, senders push content into the network in an open-loop manner to take advantage of underutilised links. When content hits the bottleneck link it gets re-routed through alternative uncongested paths. If alternative paths do not exist, incoming content is temporarily stored in in-network caches, while the system enters a closed-loop, back-pressure mode of operation to avoid congestive collapse.

The proposed shift from host-centric to information-centric networking (ICN) has triggered extensive research in the area of content naming. Efforts have so far focused on the scalability and security properties that can make content objects routable and self-certifying. In this paper, we argue that the information that is exposed through explicitly naming content objects has been overlooked, although several operational and performance issues depend on the information that a name holds. We therefore revisit content naming design decisions taking into account information exposure and deployability of the ICN paradigm.

Content Centric Networking (CCN) is a recently proposed information-centric Internet architecture in which the main network abstraction is represented by location-agnostic content identifiers instead of node identifiers. In CCN each content object is divided into packet-size chunks. When a content object is transferred, routers on the path can cache single chunks which they can use to serve subsequent requests from other users. Since content chunks in CCN may be retrieved from a number of different nodes/caches, implicit-feedback transport protocols will not be able to work efficiently, because it is not possible to set an appropriate timeout value based on RTT estimations given that the data source may change frequently during a flow. In order to address this problem, we propose in this paper a scalable, implicit-feedback congestion control protocol, capable of coping with RTT unpredictability using a novel anticipated interests mechanism to predict the location of chunks before they are actually served. Our evaluation shows that our protocol outperforms similar receiver-driven protocols, in particular when content chunks are scattered across network paths due to reduced cache sizes, long-tail content popularity distribution or the adoption of specific caching policies.