CloudLab Overview
1 like1,101 views
CloudLab is a flexible and scalable scientific infrastructure designed for transformative cloud research, enabling users to build their own clouds on provided hardware resources. It supports various research areas, including cloud architecture, big data frameworks, and multi-tenant isolation models, while being accessible to U.S. academics and educators at no charge. The initiative is in active development and aims for public availability by early 2015, with applications in critical fields such as medicine and emergency response.
CloudLab Overview
- 2. CloudLab updated: 8/14/14 The Need Addressed by CloudLab 2 • Clouds are changing the way we look at a lot of problems • Giving us new ideas of what’s possible • Impact goes far beyond computer science • … and have broader impacts with much more potential • Transformational for IT-‐‑based businesses – enables rapid startup • … but there'ʹs still a lot we don'ʹt know, from perspective of • Researchers (those who will transform the cloud) • Users (those who will use the cloud to transform their own fields) • To investigate these questions, we need: • Flexible, scalable scientific infrastructure • That enables exploration of fundamental science in the cloud • Built by and for the research community
- 3. CloudLab updated: 8/14/14 The CloudLab Vision • A “meta-‐‑cloud” for building clouds • Build your own cloud on our hardware resources • Agnostic to specific cloud software • Run existing cloud software stacks (like OpenStack, Hadoop, etc.) • … or new ones built from the ground up • Control and visibility all the way to the bare metal • “Sliceable” for multiple, isolated experiments at once 3 With CloudLab, it will be as easy to get a cloud tomorrow as it is to get a VM today
- 4. CloudLab updated: 8/14/14 What Is CloudLab? Slice B Stock OpenStack Slice A Geo-Distributed Storage Research • Supports transformative cloud research • Built on Emulab and GENI • Control to the bare metal • Diverse, distributed resources • Repeatable and scientific Slice D Allocation and Scheduling Research for Cyber-Physical Systems Slice C Virtualization and Isolation Research Utah Wisconsin Clemson GENI CC-NIE, Internet2 AL2S, Regionals 4
- 5. CloudLab updated: 8/14/14 CloudLab’s Hardware One facility, one account, three locations Wisconsin Clemson Utah • Storage and net. • Per node: • 128 GB RAM • 2x1TB Disk • 400 GB SSD • Clos fat-‐‑tree • Cisco • High-‐‑memory • 16 GB RAM / core • 16 cores / node • Bulk block store • Net. up to 56Gb • High capacity • Dell • Power-‐‑efficient • ARM64 / x86 • Power monitors • Flash on ARMs • Disk on x86 • Very dense • HP 5 • About 5,000 cores each (15,000 total) • 8-‐‑16 cores per node • Baseline: 4GB RAM / node • Latest virtualization hardware • TOR / Core switching design • 10 Gb to nodes, SDN • 100 Gb to Internet2 AL2S • Partnerships with multiple vendors
- 6. CloudLab updated: 8/14/14 Technology Foundations • Built on Emulab and GENI (“ProtoGENI”) • In active development at Utah since 1999 • Several thousand users (incl. GENI users) • Provisions, then gets out of the way • “Run-‐‑time” services are optional • Controllable through a web interface and GENI APIs • Scientific instrument for repeatable research • Physical isolation for most resources • Profiles capture everything needed for experiments • Software, data, and hardware details • Can be shared and published (eg. in papers) 6
- 7. CloudLab updated: 8/14/14 Who can use CloudLab? • US academics and educators • Researchers in cloud architecture and novel cloud applications • Teaching classes, other training activities • No charge: free for research and educational use • International federations expected • Apply on the website at www.cloudlab.us 7
- 8. CloudLab updated: 8/14/14 Early Interest in CloudLab Dorian Arnold (New Mexico) Andreas Mauthe (Lancaster) 8
- 9. CloudLab updated: 8/14/14 Cloud Architecture Research • Exploring emerging and extreme cloud architectures • Evaluating design choices that exercise hardware and software capabilities • Studying geo-‐‑distributed data centers for low-‐‑latency applications • Developing different isolation models among tenants • Quantifying resilience properties of architectures • Developing new diagnostic frameworks • Exploring cloud architectures for cyber-‐‑physical systems • Enabling realtime and near-‐‑realtime compute services • Enabling data-‐‑intensive computing (“big data”) at high performance in the cloud 9
- 10. CloudLab updated: 8/14/14 Application Research Questions 10 • Experiment with resource allocation and scheduling • Develop enhancements to big data frameworks • Intra-‐‑ and inter-‐‑datacenter traffic engineering and routing • New tenant-‐‑facing abstractions • New mechanisms in support of cloud-‐‑based services • Study adapting next-‐‑generation stacks to clouds • New troubleshooting and anomaly detection frameworks • Explore different degrees of security and isolation • Composing services from heterogeneous clouds • Application-‐‑driven cloud architectures
- 11. CloudLab updated: 8/14/14 Federated with GENI • CloudLab can be used with a GENI account, and vice-‐‑versa • GENI Racks: ~ 50 small clusters around the country • Programmable wide-‐‑area network • Openflow at dozens of sites • Connected in one layer 2 domain • Large clusters (100s of nodes) at several sites • Wireless and mobile • WiMax at 8 institutions • LTE / EPC testbed (“PhantomNet”) at Utah • International partners • Europe (FIRE), Brazil, Japan 11
- 12. CloudLab updated: 8/14/14 Community Outreach • Applications in areas of national priority • Medicine, emergency response, smart grids, etc. • Through • “Opt in” to compute jobs from domain scientists • Summer camps • Through Clemson data-‐‑intensive computing program • Under-‐‑represented groups 12
- 13. CloudLab updated: 8/14/14 Availability and Schedule • Availability: • Now (summer 2014): Technology preview available! • Late 2014: Open to early adopters • Early 2015: Publicly available • Hardware being deployed in stages: • Fall 2014: Dell / Clemson cluster • Winter 2014: Wisconsin / Cisco cluster • Spring 2015: Utah / HP cluster • Hardware refreshes in 2015 and 2016 13
- 14. CloudLab updated: 8/14/14 The CloudLab Team Robert Ricci (PI) Eric Eide Steve Corbató Kobus Van der Merwe Aditya Akella (co-‐‑PI) Remzi Arpaci-‐‑Dusseau Miron Livny KC Wang (co-‐‑PI) Jim Bonum Jim Pepin Amy Apon Chip Ellion (co-‐‑PI) Larry Landweber Mike Zink (co-‐‑PI) David Irwin 14 Glenn Ricart (co-‐‑PI)
- 15. CloudLab updated: 8/14/14 Learn more, sign up: 15 www.CloudLab.us This material is based upon work supported by the National Science Foundation under Grant No. 1419199. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.