|Schäfer, David Richard; Tariq, Muhammad Adnan; Rothermel, Kurt: Highly Available Process Executions. |
Universität Stuttgart, Fakultät Informatik, Elektrotechnik und Informationstechnik, Technischer Bericht Informatik Nr. 2016/02.
11 Seiten, englisch.
|CR-Klassif.||C.2.4 (Distributed Systems)|
C.4 (Performance of Systems)
|Keywords||processes; availability; robustness; failure tolerance; consistency; workflows|
Modern businesses operate globally with business locations and partners scattered across the globe. To handle the complexity arising with this distribution, automated business processes help managing the data and interactions between locations and partners. This migrates the execution of business processes, which previously ran in reliable back-end systems, to massively distributed environments. Due to heterogeneity and high dynamics, a key challenge in such environments is to ensure availability for these process executions in the presence of frequent communication and device failures. In this paper, we propose a protocol that provides high availability by replicating process executions. To ensure that a replicated execution produces the same outcome as a non-replicated execution, we formally define one-copy consistency for replicated process executions. Subsequently, we define rules that ensure one-copy consistency and present a replication protocol adhering to these rules even in the presence of failures. Our evaluations on Amazon EC2 and PlanetLab show that our proposed protocol significantly increases the availability of processes executing in a distributed environment while incurring very low overhead in terms of bandwidth utilization and execution time.
|PDF (993104 Bytes)|
|Abteilung(en)||Universität Stuttgart, Institut für Parallele und Verteilte Systeme, Verteilte Systeme|
|Eingabedatum||7. März 2016|