Talks

Existing research in computational storage primarily explores pushdown of a single query operator or its supporting functions from a database management system. In these cases, the portion of a query that can be pushed down is static, the execution of the kernel is device-specific, and the benefits of computational storage only exist between the database system and the compuational storage devices (CS devices) it is directly connected to.

We have seen significant growth in data and data processing applications in recent years. For such applications, data movement has been one of the primary bottlenecks for performance and Energy consumption.

There are at least two methods for enabling computational storage workflows on embedded storage hardware. The first revolutionary path is to port applications to relatively lightweight storage abstractions that allow simple software stacks.

Simple enough to describe in one sentence, the theory of operation of XPI, an interface that targets acceleration starting to show up in and near memory and storage devices, revolves around three core concepts.

Advanced, scientific-computing workflows rely on composable data service libraries to migrate data between different simulation and analysis jobs that run in parallel on a high-performance computing (HPC) platform. While these services are essential for staging and transforming in-transit data, they consume compute node resources that could otherwise be used for scientific work.

Open source software communities and their strategies and techniques has immense potential to amplify the impact of experimental systems research. The Center for Research in Open Source Software (cross.ucsc.edu) since I founded it in 2015 has successfully demonstrated this potential by raising $2.

High-performance computing (HPC) systems researchers have proposed using current, programmable network interface cards (or SmartNICs) to offload data management services that would otherwise consume host processor cycles in a platform. While this work has successfully mapped data pipelines to a collection of SmartNICs, users require a flexible means of inspecting in-transit data to assess the live state of the system.

Remote procedure calls are a major contributor to performance variance in distributed systems due to lack of isolation and contention on shared resources. In this talk, I will introduce the novel architecture we built to address this problem, and cover our experience in implementing a prototype over gRPC with an emphasis on the communication layer between a host CPU and the SmartNIC.

The amount of data stored in data centers worldwide is increasing faster than ever. Much of this data is stored to process, analyze, and extract valuable insights. Primarily, large datasets are stored in dumb storage servers inside data centers.

High-performance computing data centers supporting large-scale simulation applications can routinely generate a tremendous amount of data. For cost-effective high bandwidth, many data centers have used tiered storage with warmer tiers made of flashes and cooler tiers provisioned with high-density rotational disk drives.