The RIVER (
Research Infrastructure to explore Volatility, Energy-efficiency, and Resilience) system is a shared testbed with a dynamic range of power, performance, and reliability, enabling computer science researchers to explore the properties of future systems at realistic scale and create novel solutions that addresses performance, efficiency, resilience, and elasticity at every level of abstraction.

RIVER enables intelligent control of hardware system properties to achieve an application objective, as well as flexible response by software and applications to imposed hardware properties.

The RIVER system is available to Computer Science researchers at UChicago, but particularly those in the Department of Computer Science.

Example Research Experiments involve:

(1)  Large dynamic range of compute, memory, storage (Gigaops to Teraops, 10W to 10kW, GB to 10TB DRAM)
(2)  Active data, Big Data Analysis (10TB DRAM, 60TB SSD, customizable drives)
(3)  Power Monitoring and Control (power capping, fine-grain monitoring)
(4)  Resilience: varied fault injection
(5)  Volatility: Intermittent computing, networking and storage
(6)  ... and more...

RIVER supports a range of standard usage models (batch, 1-node big data, interactive and low-level experiments)

Hardware Configuration

Each RIVER compute nodes consists of:

  • CPU: 2 x Intel Xeon E2650 v3 12-core processor, 2.3GHz, 30MB cache, 24 cores total
  • DRAM: 16 x 16GB TruDDR4 Memory 2133MHz, 256GB total
  • Disks: 2 x 800GB SATA MLC SSD, 1.6TB total
  • Power Management: The Haswell processor supports software controllable power management including independent power budgeting for cores, graphics processors, and DRAM.
  • Network: one 10 GigE Ethernet NIC + one 1GigE Ethernet NIC