Specialized Systems: Blending Hardware and Software (in English)

## Details

In this meetup, we dive into specialized systems that are not purely software or hardware-based but focus on systems and solutions that include a unique combination of architecture and algorithms. Such solutions may change the boundaries between hardware and software, and introduce new components into the software/hardware stack that together can optimize overall system performance. We bring two examples from operating systems and storage, to highlight work that goes beyond mundane computer science and includes a fusion of hardware and software.

In collaboration with SYSTOR 2022 - https://www.systor.org/2022/

Agenda:
(14:30) Gathering, registration, mingling
(15:00) Prof. Mark Silberstein, Technion: Accelerator-centric distributed OS for heterogeneous servers and data-centers
(15:45) Refreshments
(16:00) Edward Bortnikov, Ph.D - VP Technology, Pliops: Key-Value Storage Revolutionized with Hardware Accelerated Storage Engine
Session #1: Accelerator-centric distributed OS for heterogeneous servers and data-centers / Prof. Mark Silberstein
Future systems will be omni-programmable: alongside CPUs, GPUs, Security accelerators and FPGAs, they will execute user code near-storage, near-network, and near-memory. Ironically, while breaking power and memory walls via hardware specialization and near data processing, the emerging programmability wall will become a key impediment for materializing the promised performance and power
efficiency benefits of omni-programmable systems. I argue that the root cause of the programming complexity lies in today's CPU-centric operating system (OS) design which is no longer appropriate for omni-programmable systems. In this talk I will describe the ongoing efforts in my lab to build an accelerator-centric OS called OmniX, which extends standard OS abstractions into accelerators while maintaining a coherent view of the system among all the processors. In OmniX, near-data computation accelerators may directly invoke tasks and access I/O services among themselves, excluding the CPU from the performance-critical data and control plane operations, and turning the CPU into a “yet another” accelerator for sequential computations.
I will show how the OmniX OS enables high efficiency at a level of a single server. Further, I will demonstrate that the OmniX design principles are future-proof: they are equally applicable to a large-scale disaggregated data center environment as we show via our recent data-center scale operating system called FractOS.
Session #2: Key-Value Storage Revolutionized with Hardware Accelerated Storage Engine/ Edward Bortnikov, Ph.D
Multiple modern data management platforms have at their core a key-value storage engine -- a performance-critical component that abstracts away the physical storage properties and provides access to the data as a collection of key-value pairs. In recent years, storage engine technologies came under a spotlight as the SSD speed skyrocketed and the traditionally fast CPU's emerged as the new bottleneck. We present XDP(TM) (Xtreme Data Processor) - a specialized hardware platform designed by Pliops that accelerates key-value access to SSD drives by an order of magnitude. Applications tap into XDP through the XDP Rocks (TM) C++ library, which is API-compatible with RocksDB - a popular open-source storage engine. We demonstrate the XDP Rocks decisive performance gains versus its software baseline, through microbenchmarks as well as application-level measurements.