Cornell Systems Lunch

Decision makers need timely and accurate data to make critical decisions. In many cases, that data is presented without evidence that it is trustworthy, leaving decision makers no choice but to blindly trust that the data is correct. Data provenance is a vital technology that provides decision makers with the evidence they need to make decisions based on good data, by tracking the history of ownership and processing of data as it moves through a system. In this talk, we describe the challenges of building resilient systems that leverage data provenance to protect data, and the security guarantees necessary for trustworthy whole-system provenance. We detail our experiences with integrating data provenance into a prototype system and describe our open-source Linux Provenance Modules framework for capturing trustworthy whole-system data provenance.

Abstract: Sharding -- stateful distributed scaling -- is a fundamental building block of large-scale applications, but most have their own custom, ad-hoc implementations. Our goal is to make sharding as easily reusable as a filesystem or lock manager. Slicer is Google’s general purpose sharding service. It monitors signals such as load hotspots and server health to dynamically shard work over a set of servers. Its goals are to maintain high availability and reduce load imbalance while minimizing churn from moved work.

In this paper, we describe Slicer’s design and implementation. Slicer has the consistency and global optimization of a centralized sharder while approaching the high availability, scalability, and low latency of systems that make local decisions. It achieves this by separating concerns: a reliable data plane forwards requests, and a smart control plane makes load-balancing decisions off the critical path. Slicer’s small but powerful API has proven useful and easy to adopt in dozens of Google applications. It is used to allocate resources for web service front-ends, coalesce writes to increase storage bandwidth, and increase the efficiency of a web cache. It currently handles 2-7M req/s of production traffic. The median production Slicer-managed workload uses 63% fewer resources than it would with static sharding.

Bio: Jon Howell is a distributed systems engineer at Google working on Slicer. Previously, at Microsoft Research, he contributed to the Ironfleet mechanically verified distributed system, the Pinocchio remote verification system, the Embassies secure application architecture, the record-breaking Flat Datacenter Storage system, and the FARSITE peer-to-peer file system.

Abstract: The immense bandwidth requirements of next-generation data center networks provide an opportunity for optical circuit switching to deliver substantial cost and energy advantages over state-of-the-art electronic packet switching. However, existing optical switching proposals face deployment challenges in the data center environment due to limited scalability at the physical layer and/or significant control complexity at the network layer. Our work attempts to overcome these limitations by taking a holistic design approach where we jointly redesign the network architecture and optical switching hardware subject to data center-specific constraints. The end result is a hybrid optical-electronic network solution that delivers higher throughput using less power than a comparably-priced electronic network while remaining practical to deploy and operate at scale.

Bio: William (Max) Mellette is a postdoctoral researcher in Computer Science at UC San Diego working with George Porter on data center network architecture. He received his PhD from UC San Diego in Electrical and Computer Engineering with a concentration in Photonics, where he designed and built optical switches targeting data center networks. His research interests span all aspects of networked systems, with a focus on the intersection of hardware and network design.

Abstract: In today"s data-driven world, memory system performance is often critical to overall system performance. In this talk, we present the Hawkeye Cache, a novel method of solving the age-old problem of cache replacement. We then step back and discuss our broader vision, explaining how machine learning algorithms can help us improve the Hawkeye Cache as well as other memory system optimizations.

Bio: Calvin Lin is a University Distinguished Teaching Professor of Computer Science at The University of Texas at Austin. His research interests are in compilers, security, and computer architecture. He is also the Director of the department"s Turing Scholars Honors Program and has led an NSF-funded effort to develop and dessiminate a new high school CS Principles course that uses a project-based learning pedagogy. When he is not working, he can be found chasing his two young sons or coaching the UT men"s ultimate frisbee team.