Exploring Design Alternatives for RAMP Transactions Through Statistical Model Checking

Abstract

Abstract. In this paper we explore and extend the design space of the recent RAMP (Read Atomic Multi-Partition) transaction system for large-scale partitioned data stores. Arriving at a mature distributed system design through implementation and experimental validation is a labor-intensive task, so that only a limited number of design alternatives can be explored in practice. The developers of RAMP did implement and validate three design alternatives for RAMP, and sketched three additional designs. This work addresses two questions: (1) How can the design space of a distributed transaction system such as RAMP be explored with modest effort, so that substantial knowledge about design alternatives can be gained before designs are implemented? and (2) How realistic and informative are the results of such design explorations? We answer the first question by: (i) formally modeling eight RAMP-like designs (five by the RAMP developers and three of our own) in Maude as probabilistic rewrite theories, and (ii) using statistical model checking of those models to analyze key performance metrics such as throughput, average latency, and degrees of strong consistency and read atomicity. We answer the second question by showing that our quantitative analyses: (i) are consistent with the experimental results obtained by the RAMP developers for their implemented designs; (ii) confirm the conjectures made by the RAMP developers for their other three unimplemented designs; and (iii) uncover some promising new designs that seem attractive for some applications.

Read atomic transactions with prevention of lost updates: ROLA and its formal analysis

“…Our next planned step is to implement ROLA, evaluate it experimentally, and compare the experimental results with the formal analysis ones. In previous work on existing systems such as Cassandra [17], RAMP [7], and Walter [34], the performance estimates obtained by formal analysis and those obtained by experimenting with the real system were basically in agreement with each other [19,21,24]. This confirmed the useful predictive power of the formal analyses.…”

“…Maude and PVeStA have been used to model and analyze the correctness and performance of a number of distributed data stores: the Cassandra key-value store [25,19,20], different versions of RAMP [22,21], Walter [24], P-Store [30], and Google's Megastore [14,15]. In contrast to these papers, our paper uses formal methods to develop and validate an entirely new design, ROLA, for a new consistency model.…”

Automatic Analysis of Consistency Properties of Distributed Transaction Systems in Maude

“…These requirements are not very strict. The Maude models of the DTSs RAMP [29], Faster [24], Walter [26], ROLA [25], Jessy [28], and P-Store [32] can all be seen as instantiations of our modeling framework, with very small syntactic changes, such as defining transaction and replica objects as subclasses of Txn and Replica, changing the names of the attributes and sorts, etc. The Apache Cassandra NoSQL key-value store can be seen as a transaction system where each transaction is a single operation; the Maude model of Cassandra in [30] can also be easily modified to fit within our modeling framework.…”

Design and Validation of Cloud Storage Systems Using Formal Methods

“…Having higher-level smaller formal models allowed us to explore the design state of RAMP quite extensively. In particular, in [10] we used statistical model checking to evaluate the performance along a number of parameters, with many different distributions of transactions. In this way, we could evaluate the performance of a number RAMP designs not explored by Bailis et al, and for many more parameters and workloads than evaluated by the RAMP developers.…”