Machine virtualization has become a cornerstone of modern datacenters. It enables server consolidation as a means to reduce costs and increase efficiencies. The communication endpoints within the datacenter are now virtual machines (VMs), not physical servers. Consequently, the datacenter network now extends into the server and last hop switching occurs inside the server. Today, thanks to increasing core counts on processors, server VM densities are on the rise. This trend is placing enormous pressure on the network I/O subsystem and the last hop virtual switch to support efficient communication, both internal and external to the server. But the current state-of-the-art solutions fall short of these requirements. This thesis presents new architectures and mechanisms for the network subsystem in virtualized servers to build efficient virtualization platforms. Specifically, there are three primary contributions in this thesis. First, it presents a new mechanism to reduce memory sharing overheads in driver domain-based I/O architectures. The key idea is to enable a guest operating system to reuse its I/O buffers that are shared with a driver domain. Second, it describes Hyper-Switch, a highly streamlined, efficient, and scalable software-based virtual switching architecture, specifically for hypervisors that support driver domains. The Hyper-Switch combines the best of the existing architectures by hosting the device drivers in a driver domain to isolate any faults and placing the virtual switch in the hypervisor to perform efficient packet switching. Further, the Hyper-Switch implements several optimizations, such as virtual machine state-aware batching, preemptive copying, and dynamic offloading of packet processing to idle CPU cores, to enable efficient packet processing, better utilization of the available CPU resources, and higher concurrency. This architecture eliminates the memory sharing overheads associated with driver domains. Third, this thesis proposes an alternate virtual switching architecture, called sNICh, which explores the idea of server/switch integration. The sNICh is a combined network interface card (NIC) and datacenter switching accelerator. This takes the Hyper-Switch architecture one step further. It offloads the data plane of the switch to the network device, eliminating driver domains entirely.

Multi-core technology is bringing parallel processing capabilities from servers to laptops and even handheld devices. At the same time, platform support for system virtualization is making it easier to consolidate server and client resources, when and as needed by applications. This consolidation is achieved by dynamically mapping the virtual machines on which applications run to underlying physical machines and their processing cores. Low cost processor and I/O virtualization methods efficiently scaled to different numbers of processing cores and I/O devices are key enablers of such consolidation. This dissertation develops and evaluates new methods for scaling virtualization functionality to multi-core and future many-core systems. Specifically, it re-architects virtualization functionality to improve scalability and better exploit multi-core system resources. Results from this work include a self-virtualized I/O abstraction, which virtualizes I/O so as to flexibly use different platforms' processing and I/O resources. Flexibility affords improved performance and resource usage and most importantly, better scalability than that offered by current I/O virtualization solutions. Further, by describing system virtualization as a service provided to virtual machines and the underlying computing platform, this service can be enhanced to provide new and innovative functionality. For example, a virtual device may provide obfuscated data to guest operating systems to maintain data privacy; it could mask differences in device APIs or properties to deal with heterogeneous underlying resources; or it could control access to data based on the ``trust' properties of the guest VM. This thesis demonstrates that extended virtualization services are superior to existing operating system or user-level implementations of such functionality, for multiple reasons. First, this solution technique makes more efficient use of key performance-limiting resource in multi-core systems, which are memory and I/O bandwidth. Second, this solution technique better exploits the parallelism inherent in multi-core architectures and exhibits good scalability properties, in part because at the hypervisor level, there is greater control in precisely which and how resources are used to realize extended virtualization services. Improved control over resource usage makes it possible to provide value-added functionalities for both guest VMs and the platform. Specific instances of virtualization services described in this thesis are the network virtualization service that exploits heterogeneous processing cores, a storage virtualization service that provides location transparent access to block devices by extending the functionality provided by network virtualization service, a multimedia virtualization service that allows efficient media device sharing based on semantic information, and an object-based storage service with enhanced access control.