Contributions to Content Placement, Load-balancing and Caching: System Design and Security

Distributed systems invaded our daily lives offering a large set of services and applications such as social networks, chats, video consumption etc. Despite the large diversity of services offered, the underlying systems have many things in common. They must handle a large set of computing and storage nodes and must decide where to place the data and how to balance the load between the nodes. Therefore, these systems implement some placement and load-balancing policies. Data caches are used to further optimize access to data. The caching policy - a specific instance of the placement policy - decides which items to keep in cache and which ones to evict. Even small scale and dedicated architectures such as hard disk drives or CPUs employ the above policies and mechanisms.This document presents several contributions to content placement, load-balancing and caching policies. We consider two different aspects: (i) the system design, i.e. the design and evaluation of such policies for different architectures and (ii) the security, i.e. understanding how to exploit the caches, either for an attack or to build security applications.We start by presenting several contributions around the system design of such policies for Video-on-Demand services. We first consider a system where the video content is pro-actively placed on home gateways managed by an operator. During content consumption the gateways fetch the content from other home gateways. In contrast, the second system we study exclusively exploits the cloud. It serves targeted videos and must decide which content to cache instead of recomputing it at each user request. The pay-per use model of the cloud requires new and specific attention when designing the caching policies. We then consider security applications and attacks of caches. We build a trusted time stamping service using the caches of the publicly available Domain Name Service. We also design and demonstrate attacks on the caches of modern CPUs, allowing us to pass information from one process (or virtual machine) to another despite the isolation. Finally, we conclude with an outlook and some future work in the domain of caching and load-balancing