@article{plagge_feinberg_mcfarland_rothganger_agarwal_awad_hughes_cardwell_2022, title={ATHENA: Enabling Codesign for Next-Generation AI/ML Architectures}, DOI={10.1109/ICRC57508.2022.00016}, abstractNote={There is a growing market for technologies ded-icated to accelerating Artificial Intelligence (AI) workloads. Many of these emerging architectures promise to provide savings in energy efficiency, area, and latency when compared to traditional CPUs for these types of applications. In particular, neuromorphic analog and digital technologies provide both low-power and configurable acceleration of challenging artificial intelligence (AI) algorithms. If designed into a heterogeneous system with other accelerators and conventional compute nodes, these technologies have the potential to augment the capabilities of traditional High Performance Computing (HPC) platforms. We present a codesign ecosystem that leverages an analytical tool, ATHENA, to accelerate design space exploration and evaluation of novel architectures.}, journal={2022 IEEE INTERNATIONAL CONFERENCE ON REBOOTING COMPUTING, ICRC}, author={Plagge, Mark and Feinberg, Ben and McFarland, John and Rothganger, Fred and Agarwal, Sapan and Awad, Amro and Hughes, Clayton and Cardwell, Suma G.}, year={2022}, pages={13–23} }
 @article{mcfarland_awad_2022, title={Transpose-Xen: Virtualized Mixed-Criticality through Dynamic Allocation}, DOI={10.1145/3477314.3506979}, abstractNote={Cloud systems continue to rise in popularity due to their ability to provide access to flexible, scalable systems to be shared among all their users. Several tasks can be executed simultaneously within a server, but have varying requirements for completion. While some jobs may have latency-critical quality-of-service (QoS) requirements, others may have stricter real-time constraints for maximum deadline misses. The introduction of hard real-time tasks, where zero deadline misses are acceptable, results in scheduling concurrent jobs becoming increasingly difficult. In this paper we propose Transpose-Xen, an adaptive hyper-visor scheduler capable of managing the scheduling of tasks of varying levels of criticality, including tasks with hard real-time constraints. Transpose-Xen is able to execute multiple jobs of varying criticality by finding similarities between the resource needs of each task despite potentially executing from separate VMs. Once grouped into these resource sub-pools, or ponds, our scheduler allocates the needed resources to ensure that each job is schedulable if possible. Transpose-Xen also leverages the use of virtual-deadlines, a scheduling algorithm that we use to prioritize higher-criticality tasks without completely starving lower-criticality tasks of resources. By profiling the impact of resource allocation on real-time tasks, different jobs of varying criticality levels can be scheduled concurrently - capable of satisfying both hard real-time constraints and satisfying up to 99% of all soft real-time deadlines.}, journal={37TH ANNUAL ACM SYMPOSIUM ON APPLIED COMPUTING}, author={McFarland, John and Awad, Amro}, year={2022}, pages={3–12} }