Keep net working - on a dependable and fast networking stack

Abstract

For many years, multiserver 1 operating systems have been demonstrating, by their design, high dependability and reliability. However, the design has inherent performance implications which were not easy to overcome. Until now the context switching and kernel involvement in the message passing was the performance bottleneck for such systems to get broader acceptance beyond niche domains. In contrast to other areas of software development where fitting the software to the parallelism is difficult, the new multicore hardware is a great match for the multiserver systems. We can run individual servers on different cores. This opens more room for further decomposition of the existing servers and thus improving dependability and live-updatability. We discuss in general the implications for the multiserver systems design and cover in detail the implementation and evaluation of a more dependable networking stack. We split the single stack into multiple servers which run on dedicated cores and communicate without kernel involvement. We think that the performance problems that have dogged multiserver operating systems since their inception should be reconsidered: it is possible to make multiserver systems fast on multicores.

Shrinking the hypervisor one subsystem at a time

“…Hruby et al [12] agree that dedicating cores to I/O functionality is worthwhile and rearchitect a TCP/IP stack by putting the different layers onto different cores. In contrast to sv3, which follows the run-to-completion philosophy, each networking stack layer is a single threaded program.…”

Towards a Flexible, Lightweight Virtualization Alternative

“…In a straightforward implementation, this would several extra context switches compared to the other architectures. Future research will have to reveal to which extent any resulting overheads can be alleviated, for example with a small domain-local cache, memory mapping based approaches, or asynchronous operations [24,41,47]. For the host system, the resulting size of the trusted computing base (TCB) will be somewhere between that of hardwarelevel and OS-level virtualization.…”