A Crash-Tolerant Proxy Service for High-Availability Environments

BBB-Link: https://bbb.tu-dresden.de/b/mat-xin-oyh-xzn

Presentation Language: German

The central role of proxies in networks places them in a critical position, where any interruption can significantly impact adjacent services. Therefore, their ability to recover from faults can become an important requirement of the service.
This work investigates crash-tolerance mechanisms of such a proxy service, entirely implemented in userspace. By employing redundant processes, replication of file descriptors, and shared memory, the resulting proxy can recover from unexpected process terminations, as long as one of its processes persists. Operations on the central shared memory, including socket I/O, are performed through transactions, so that it remains in a consistent state at all times.
The effectiveness of the approach is evaluated qualitatively and quantitatively. Connection persistence is achieved and thousands of concurrent connections can be recovered in less than a second. When compared to a similar architecture, memory overhead is minimal but computational overhead of transactional I/O and memory access can vary widely, depending on the workload. However, when compared to existing conventional proxies, further improvements are necessary to achieve the same level of performance.