Logo

Header

Header

Mark Silberstein

Technion

OmniX - an OS Architecture for Omni-Programmable Systems

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, 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 todays CPU-centric operating system (OS) design which is no longer appropriate for omni-programmable systems.

In this talk I will summarize the results of the last seven years of research and a handful of publications on an accelerator-centric OS architecture called OmniX. OmniX advocates for extending standard OS abstractions into accelerators, while maintaining a coherent view of the system among all the resident processors. In OmniX, near-data and compute accelerators may directly invoke tasks and access I/O services, excluding the CPU from performance-critical data and control plane operations.The CPU becomes a "yet another" accelerator for sequential computations. I will show how OmniX design principles have been successfully applied to GPUs, Programmable NICs, and Intel SGX, as well as disaggregated cloud systems, and discuss the obstacles and successes of adopting some of them in industry.

Photo
Mark Silberstein is an Associate Professor at the EE department at the Technion - Israel Institute of Technology where he is heading the Accelerated Computing Systems Lab. His research is centered around OSes for compute and I/O accelerator architectures. He is also working on practical ways to protect systems from speculative and non-speculative side channels attacks, showing some new attacks (Foreshadow) but also providing efficient ways to detect existing vulnerabilities (SpecFuzz, USENIX Sec'20). His recent new passion is using neural nets as a "computational" cache, where his student demonstrated the first application of NNs to data-path packet classification (NuevoMatch, SIGCOMM'20).


Homepage: https://marksilberstein.com