Project Overview
The Hydro project is a framework for data-centric cloud programming research from the RISE Lab at UC Berkeley. Our work thus far has focused on developing low-latency systems that support coordination-free consistency techniques to resolve potentially conflicting updates to shared state. The two systems we have developed thus far are Anna and Cloudburst.
Anna
Anna is a low-latency autoscaling KVS that aims to provide the latency guarantees of a memory-speed KVS like Redis while also autoscaling to match workload, similar to systems like AWS S3 or Google Cloud Storage. To avoid expensive locking or lock-free atomic techniques, Anna allows each worker thread to accept updates to the same keys in parallel without coordination. To resolve conflicting updates, Anna stores all data in lattice data structures which have associate, commutative, and idempotent merge functions.
Cloudburst
Cloudburst is a stateful Function-as-a-Service (FaaS) platform built on top of Anna. Cloudburst focuses on providing best-in-class performance for FaaS workloads that share state (e.g., function composition, direct communication) by introducing caches that live on the same physical machines as function executors. Cloudburst also guarantees causal consistency and repeatable reads for compositions of functions by leveraging Anna's consistency primitives and adding new consistency protocols at the caching layer.
Research Papers
- New Directions in Cloud Programming. A. Cheung, N. Crooks, J. M. Hellerstein, M. Milano. CIDR. 2021.
- Optimizing Prediction Serving on Low-Latency Serverless Dataflow. V. Sreekanti, H. Subbaraj, C. Wu, J. E. Gonzalez, J. M. Hellerstein. arxiv:2007.05832. 2020.
- Cloudburst: Stateful Functions-as-a-Service. V. Sreekanti, C. Wu, X. C. Lin, J. Schleier-Smith, J. M. Faleiro, J. E. Gonzalez, J. M. Hellerstein, A. Tumanov. PVLDB. 2020.
- Transactional Causal Consistency for Serverless Computing. C. Wu, V. Sreekanti, J. M. Hellerstein. SIGMOD 2020.
- A Fault-Tolerance Shim for Serverless Computing. V. Sreekanti, C. Wu, S. Chhatrapati, J. E. Gonzalez, J. M. Hellerstein, J. M. Faleiro. EuroSys 2020. To appear.
- Autoscaling Tiered Cloud Storage in Anna. C. Wu, V. Sreekanti, J.M. Hellerstein. VLDB 2019. (Preprint: arXiv:1809.00089.)
- Serverless Computing: One Step Forward, Two Steps Back. J.M. Hellerstein, J. Faleiro, J.E. Gonzalez, J. Schleier-Smith, V. Sreekanti, A. Tumanov, C. Wu. CIDR 2019. (Preprint: arXiv:1812.03651.)
- Anna: A KVS for Any Scale. C. Wu, J. M. Faleiro, Y. Lin, J. M. Hellerstein. ICDE 2018, TKDE 2019.
Talks
- A Data-Centric Lens on Cloud Programming and Serverless Computing. Joe Hellerstein. Keynote, ICDE20. Darmstadt, Germany. (Slides, Video)
- Adavance Calmly: Serverless Computing & Cloud Programming. Joe Hellerstein. Distinguished Lecture, University of Wisconsin, Madison. Madison, WI, USA. (Slides)
- The State of Serverless Computing. Chenggang Wu. QCon New York. June 24, 2019. New York, USA.
- The State of Serverless Computing, or Fixing Dysfunction-as-a-Service. Vikram Sreekanti and Chenggang Wu. Craft Conference. May 9, 2019. Budapest, Hungary.
Blog Posts
- The State of the Serverless Art. Joe Hellerstein. August 13, 2020.
- Real-Time Prediction Serving, Simplified. Vikram Sreekanti. July 14, 2020.
- Solving Serverless Computing's Fault-Tolerance Problem. Vikram Sreekanti. March 19, 2020.
- Cloudburst: Stateful Function-as-a-Service. Vikram Sreekanti, Joe Hellerstein. January 28, 2020.
- Serverless Computing: One Steps Forward, Two Steps Back. Joe Hellerstein. December 13, 2018.
- Going Fast and Cheap: How We Made Anna Autoscale. Chenggang Wu, Vikram Sreekanti, Joe Hellerstein. September 7, 2018.
- Anna: A Crazy Fast, Super-Scalable, Flexibly Consistent KVS. Joe Hellerstein. March 9, 2018.