CNS Core:Medium:Collaborative Research:Towards Enabling Optimal Performance-Cost Tradeoffs in Distributed Storage

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

Modern Internet services aim to be always available and provide low latency responses. Typically, this is achieved by storing data on multiple sites close to the end-users. Such an arrangement imposes a fundamental trade-off between response times for read and write requests to storage systems and data storage/transfer costs. Existing distributed storage solutions for addressing this trade-off may result in sub-optimal outcomes. First, realizing different performance-vs-cost trade-offs today requires using radically different solutions with little choice in between. Also, many trade-off points that appear theoretically feasible are currently unachievable in practice. This research project aims to design and build a distributed storage solution that offers a single design to realize a wide range of feasible latency-cost trade-offs. The overarching goal of this project is to design next-generation distributed storage solutions (1) that can be configured to achieve all feasible performance-cost trade-offs on mutable data, and (2) that enables a significant portion of the theoretically feasible trade-off space that is currently unachievable. Specifically, the project involves overcoming the following key challenges: (1) Seamless support for low latency and low cost; (2) Ensuring high performance across widely varying network latency domains and object sizes; and (3) Efficiently maintaining consistency in erasure-coded data. This will allow application developers to select from all feasible points in the trade-off space, using replication or erasure coding, without having to redesign their services. By making a broader region of the trade-off space accessible to cloud service providers, this project will help reduce the price of cloud storage for end users. Additionally, lowering achievable latency bounds enables new, low-cost, geo-distributed services and applications that are interactive and collaborative. The researchers will work with industry partners to apply the outcomes from this project in practice, and leverage the results from this project in classes they teach. Research exposure for undergraduate students will be promoted through research internships. The researchers plan to build upon several already-established outreach activities to help improve diversity of the student population in computer science. The results from the research project, including software, will be made available at: github.com/cns1901410. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.