Search

Social networking applications' disk access patterns differ from those of traditional applications. However, today's disk layout techniques aren't adapted to social networking workloads, and thus their performance suffers. The authors' disk layout techniques leverage community structure in a social graph to make placement decisions that optimize read latency. Their layout manager, Bondhu, incorporates these techniques and is integrated into the popular Neo4j graph database engine. Experimental results show that Bondhu improves the median response time for online social network operations by as much as 48 percent.

Attribute-based Access Control (ABAC) based on XACML can substantially improve the security and management of access rights on databases. However, existing implementations rely on high-level policy interpretation and are not as efficient as mechanisms natively supported by commodity databases. In this paper we explore advantages and challenges arising from compiling XACML policies for database access into Access Control Lists (ACLs) natively supported by the database. The main contributions are an architecture and algorithms for efficiently addressing incremental changes in attributes that could trigger changes to the ACLs. We consider this in a context of reflective database access control where attributes used in access decisions are stored in the database itself. Our implementation and experiments demonstrate a significant improvement in access decision times compared to the best available optimizations for general XACML access engines.

Parallel dataflow programs generate enormous amounts of distributed data that are short-lived, yet are critical for completion of the job and for good run-time performance. We call this class of data as intermediate data. This paper is the first to address intermediate data as a first-class citizen, specifically targeting and minimizing the effect of run-time server failures on the availability of intermediate data, and thus on performance metrics such as job completion time. We propose new design techniques for a new storage system called ISS (Intermediate Storage System), implement these techniques within Hadoop, and experimentally evaluate the resulting system. Under no failure, the performance of Hadoop augmented with ISS (i.e., job completion time) turns out to be comparable to base Hadoop. Under a failure, Hadoop with ISS outperforms base Hadoop and incurs up to 18% overhead compared to base no-failure Hadoop, depending on the testbed setup.