On Fault-tolerant and High Performance Replicated Transactional Systems

(ABSTRACT) With the recent technological developments in last few decades, there is a notable shift in the way business/consumer transactions are conducted. A majority of transaction now a days fall in Online Transaction Processing (OLTP) category, where transactions are triggered over the internet and transactional systems working in the background ensure that these transactions are processed. One of the most important requirements of these OLTP transaction systems is dependability. Replication is a common technique that makes the services dependable and therefore helps in providing reliability, availability and fault-tolerance. Active replication based replicated transaction systems exploit full replication to avoid service interruption in case of node failures. Deferred Update Replication (DUR) and Deferred Execution Replication (DER) represent the two well known transaction execution models for replicated transactional systems. Under DUR, transactions are executed by clients before a global certification is invoked to resolve conflicts against other transactions running on remote nodes. On the other hand, DER postpones the transaction execution until the agreement on a common order of transaction requests is reached. Both DUR and DER require a distributed ordering layer, which ensures a total order of transactions even in case of faults. In today's distributed transaction processing systems, performance is of paramount importance. Any loss in performance results e.g. latency resulting from slow processing of a client request or slow page load, results in loss of revenue for businesses. Although DUR model perform good when conflicts are rare, it is found to be worst impacted by high conflict work-load profiles. In contrast, immunity from percentage of conflicts within transactions make DER an attractive choice of transaction execution model, but its serial execution results in limited performance and its total order layer for serializing all the transactions results in moderately high latencies. In addition to it, total order layer poses scalability challenges i.e. it does not scale with increase in system size. In this dissertation, we propose multiple innovations and system optimizations to enhance the overall performance of replicated transactional systems. First, in HiperTM we exploit the time between the instance when a client broadcasts its request and the instance when its order is finalized, to speculatively execute the request and commit it when the final order arrives. To achieve this goal, we extend S-Paxos with optimistic delivery, resulting in OS-Paxos and build a transactional system based on DER model. HiperTM uses a novel, speculative concurrency control protocol called SCC, which …