Performant and Energy Efficient Warehouse-Scale Datacenters
Improving datacenter efficiency including performance and energy efficiency is very important for data centers. In addition, increasing performance predictability and reining in the tail latency has been identified as one of the most important challenges in modern datacenter system design. To this end, we have several projects and publications that focus on this theme. We have looked at how to use fine-grained DVFS, heterogeneous architecture, design smart performance prediction capabilities to rein in the tail latency and improve datacenter efficiency.
Related Publications:
[ISCA ’15] “Thermal Time Shifting: Leveraging Phase Change Materials to Reduce Cooling Costs in Warehouse-Scale Computers”
[HPCA’15] “Adrenaline: Pinpointing and Reining in Tail Queries with Quick Voltage Boosting.”
[HPCA’15] Octopus-Man: QoS-Driven Task Management for Heterogeneous Multicore in
[MICRO ’14] SMiTe: Precise QoS Prediction on Real System SMT Processors to Improve Utilization in Warehouse Scale Computers
[USENIX ATC ’14] HaPPy: Hyperthread-aware Power Profiling Dynamically
Sirius and Djinn: Open-Source Intelligent Personal assistants and their implications on future datacenter design
If you are interested, check out the preview video on YouTube: DjiNN and Sirius
Sirius – We have designed an open-source Intelligent Personal Assistant (IPA) software, Sirius, that provides functionalities similar to Apple’s Siri, Google’s Google Now and Microsoft Cortana. We have released this software. The release has attracted much attention from the press and open-source community, including Wired, PCWorld, VentureBeat and Hacker News. The Sirius was number 1 trending project on Github for a few weeks. Using Sirius, we have conducted a study for future designs of warehouse-scale computer using various accelerators. The results are summarized in our publication at ASPLOS ’15.
Djinn – We have looked at how to provide deep learning as a general service for various IPA applications in datacenter and how that impacts future WSC designs.
Related Publications:
[ISCA ’15] “Djinn: DNN as a Service and Its Implications for Future Warehouse Scale Computers.”
[ASPLOS ’15] “Sirius: An Open End-to-End Voice and Vision Personal Assistant and Its Implications for Future Warehouse Scale Computers”
Near-free Dynamic Compiler
System adaptability for the dynamic execution environment is critical for quality of service and efficiency in data centers . To this end,we have designed and implemented a dynamic compiler that can recompile and reroute execution to code regions with very small overhead. This allows us to customize code for different execution environment to significantly increase the utilization of datacenters.
Related Publications:
[MICRO ’14] “Protean Code: Achieving Near-Free Online Code Transformations for Warehouse Scale Computers. ” Acceptance Rate: 19%