Cloud computing is emerging as a dominant computing platform for providing scalable online services to a global client base. Today's popular online services (e.g., web search, social networking, and business analytics) are characterized by massive working sets, high degrees of parallelism, and real-time constraints. These characteristics set scale-out applications apart from desktop (SPEC), parallel (PARSEC), and traditional commercial server applications. In order to stimulate research in the field of cloud and data-centric computing, we have created CloudSuite, a benchmark suite based on real-world online services. CloudSuite covers a broad range of application categories commonly found in today's datacenters. The first release includes data analytics, data serving, media streaming, large-scale and computation-intensive tasks, web search, and web serving.


Machine learning workloads exhibit computational requirements that are orders of magnitude larger than traditional datacenter workloads. To cope with such requirements, datacenter operators have to provide very efficient ML processing. The ColTraIn project co-locates DNN training and inference to increase the efficiency of datacenter DNN accelerators by harvesting the idle cycles of inference accelerators. Doing so requires DNN the redesign of inference accelerators with co-location in mind, revisiting both the datapath and the control path of DNN accelerators.


Computer architects have traditionally relied on software simulation to measure the performance metrics (e.g., instructions per cycle) of a proposed design. However, modern simulation requirements are challenging the conventional modeling tools that have traditionally served the architecture community. The QFlex project targets quick, accurate, and flexible simulation of multi-node computer systems proceeding along four fronts. The popular and open-source QEMU full-system emulator, which allows functional emulation of unmodified operating systems and applications. The powerful and flexible Flexus simulation framework, which allows detailed cycle-accurate simulation. The detailed and flexible NS-3 network simulation stack. The rigorous statistical SMARTS sampling theory to reduce the simulation turnaround time by several orders of magnitude, while achieving high accuracy and confidence estimates.

Scale-Out NUMA

Emerging datacenter applications operate on vast datasets that are kept in DRAM to minimize latency. The large number of servers needed to accommodate this massive mem- ory footprint requires frequent server-to-server communication in applications such as key-value stores and graph-based applications that rely on large irregular data structures. The fine-grained nature of the accesses is a poor match to commodity networking technologies, including RDMA, which incur delays of 10-1000x over local DRAM operations. Scale-Out NUMA is an architecture, programming model, and communication protocol for low-latency, distributed in-memory processing, designed to bridge the latency gap between local and remote memory access.

VISA: Vertically Integrated Server Architecture

The impending plateau of voltage levels with a continued increase in chip density (according to Moore's law) is causing energy to be the number one concern in the design of future digital computing platforms. These platforms are likely to be built on "dark silicon", where a limited power budget allows only for a fraction of a chip's real-estate to be active at a time, allowing for the rest of the chip to be turned off or "dark". The Vertically-Integrated Server Architecture (VISA) project targets design for dark silicon where an integrated hardware/software approach to specialization implements performance- and energy-hungry services with minimal energy. Specialization allows future technologies to utilize dark silicon effectively and maintain a constant power envelope by keeping only the needed services active on-chip, monitoring and shutting off unneeded resources. Specialization maximizes transistor efficiency and makes better use of available real-estate, achieving two or more orders of magnitude reduction in energy through a hand-in-hand collaboration of software and hardware.

© EPFL PARSA 1015 Lausanne, Switzerland       tel. +41 21 693 1395        all rights reserved       webmaster       web design by atelier.f