Back to overview

[Technical report] An overview of new features in the SPLAY framework for simple distributed systems evaluation

Type of publication Not peer-reviewed
Publikationsform Other publication (non peer-review)
Publication date 2012
Author Charles Lucas, Felber Pascal, Halalai Raluca, Rivière Etienne, Schiavoni Valerio, Valerio José,
Project MistNet: An Experimental Peer-to-peer Platform for the Cloud
Show all

Other publication (non peer-review)

Book [Technical report] An overview of new features in the SPLAY framework for simple distributed systems evaluation
Publisher Université de Neuchâtel, Neuchâtel


Evaluating large-scale distributed applications is usually a complex, time-consuming and error-prone task. The use of experimental testbeds such as PlanetLab [4], ModelNet [26]-enabled clusters, Emulab [28]), or non-dedicated testbeds such as networks of workstations, is desirable for distributed systems and algorithms evaluation but often unfavored for simulations due to the complexity gap between algorithms specifications (pseudocode) and actual imple- mentations. SPLAY is a framework that simplifies the prototyping, development, deployment and evaluation of large-scale systems. The SPLAY system and its evaluation were presented two years ago at the NSDI conference [21]. Since then, new features and possibilities have been added to it, as well as performance, scalability and flexibility improvements. This technical report relates on these evolutions, and acts as a companion report to our NSDI paper for updated information on SPLAY. Documentation and further information can be found on the project’s Web site. SPLAY is open source and can be downloaded from the same location: The novel features covered in this report include (1) user-space network topology emulation, (2) a novel distributed synchronization service, (3) improved management of churn replay from traces, (4) native library shipping for dynamically extending the libraries available on the testbed without re-deploying the persistent daemons and without accessing directly the testbed nodes, (5) improvements of the command-line interface, and (6) the possibility to use SPLAY as a batch submission service that is able to schedule and queue jobs. We conclude this technical report by some discussion of future work and directions.