Architectural Concerns in Large Datacenters

ACLD 2010 is a half-day workshop to be held in conjunction with ISCA 2010 in Saint-Malo, France on June 19. The complete workshop and tutorial schedule for ISCA 2010 can be found here. Last year's ACLD website can be found here.

Organizing committee: Christos Kozyrakis, Stanford (co-chair); Ben Sigelman, Google (co-chair); Amin Vahdat, UCSD; Partha Ranganathan, HP; Dave Andersen, CMU

Workshop format and schedule

This year, ACLD will include a keynote talk, a poster session, and two staged (but not scripted!) debates. We hope that ACLD will be something other than a mini-conference; specifically, we hope that a substantial poster session will give attendees time to interact and learn from each other. Our goal is to promote discussion and foster relationships that talks alone cannot, especially since there's a great opportunity at ISCA to get those of us who primarily think about the architecture of single systems to interact with those of us who primarily think about the architecture of distributed systems and the management of large datacenters.

2:00pm: Brief welcome and introductions
2:05pm: Keynote address
2:45pm: Debate: "Mega-servers vs. micro-blades" for datacenter workloads
3:35pm: Poster session begins (official ISCA coffee break is from 3:30->4:00; the poster session will be concurrent)
5:10pm: Debate: The future of persistent storage in the datacenter


Dan Lenoski (VP of Engineering at Cisco) will deliver the keynote address, provocatively titled as follows:

Why Cisco Should Be Your Next Server Vendor

Most modern data centers contain thousands of individual servers based on x86 processors and standard JEDEC DRAM DIMMs. Software is often supplied by a variety of independent software vendors (ISVs). Server vendors distinguish themselves via system configuration, price, I/O connectivity, management and added customer services. In these large server farms, the differentiated part of the computing infrastructure is as much in the core expertise of Cisco as any of the traditional server vendors.

This talk will outline the major features and characteristics of Cisco's Unified Computing System (UCS), a multi-chassis blade server system connected by a unified IO fabric. Emphasis will be on the unique hardware and software capabilities of UCS that enable dynamic "Software as a Service" operation and high-performance virtualization.

Dan Lenoski is Vice President of Engineering in the Server Access Virtualization Group at Cisco. Previously, he was VP of Engineering and an early member of the engineering team at the start-up Nuova Systems which Cisco acquired in 2008. From 2000 to 2005, Dan was an Engineering VP at Cisco. During this time, he led the development of the CRS-1 high-end router, initiated the efforts on Cisco's Quantum Flow network processor, and led the engineering team for Cisco's stackable switching products (Catalyst 3K and 2K). Dan was a founder and Vice President of Engineering at Growth Networks, a fab-less semiconductor company that was acquired by Cisco in 2000. Earlier, he held senior technical and management positions at Silicon Graphics, Sun Microsystems and Tandem Computers. Dan holds over twenty-five US patents in the area of network switching, processor design, multiprocessor system design, high-speed signaling and fault-tolerance; and he is co-author of the book "Scalable Shared-Memory Multiprocessing". He holds a Ph.D. and M.S. in Electrical Engineering from Stanford University, and a B.S. in Electrical Engineering from the California Institute of Technology.


Debaters / Panelists

Ben Lee
Kunle Olukotun
James Laudon
Doug Burger
Sudhanva Gurumurthi
David Anderson
The first debate concerns the appropriate architectural scale for systems which multiply to handle datacenter workloads. Our participants/panelists/debaters are Ben Lee (Stanford), Kunle Olukotun (Stanford), and James Laudon (Google), and they will be making the case for micro-, mega-, and somewhere-in-the-middle architectures respectively. (Moderated by Christos Kozyrakis (Stanford))

The second debate will focus on persistent storage in the datacenter. Doug Burger (Microsoft Research / UT Austin), Sudhanva Gurumurthi (UVA), and David Anderson (Seagate) will each argue for their vision of what large-scale persistent storage stacks will look like in the future, and whether they revolve around disk, flash, PCM, something else, or some particular combination of the above. (Moderated by Partha Ranganathan (HP))

Accepted Posters

In alphabetical order by primary author:

Performance Analysis for Large Internet Service Applications
Amer Diwan and Richard L. Sites (Google)

An Energy Saving System for Computing Clusters by Selective Activation/Deactivation of Nodes

Manuel F. Dolz, Juan C. Fernández, Rafael Mayo, Enrique S. Quintana-Ortí
Depto. de Ingeniería y Ciencia de los Computadores, Universitat Jaume I, 12071-Castellón, Spain.

Aster*x: Load-Balancing as a Network Primitive
Nikhil Handigol†, Mario Flajslik†, Srini Seetharaman‡, Ramesh Johari†, Nick McKeown† 
† Stanford University    ‡ Deutsche Telekom R&D Lab USA

The Path Forward: Specialized Computing in the Datacenter
Nikos Hardavellas*, Michael Ferdman†‡, Anastasia Ailamaki‡, Babak Falsafi‡
* Northwestern University     † Carnegie Mellon University     ‡ École Polytechnique Fédérale de Lausanne

A Case for Energy-Aware Accounting in Large-Scale Computing Facilities: Cost Metrics and Implications for Processor Design
Víctor Jiménez†, Francisco J. Cazorla†, Rober to Gioiosa†,
Eren Kursun*, Canturk Isci*, Alper Buyuktosunoglu*, Pradip Bose*, Mateo Valero†
† Barcelona Supercomputing Center, Spain        * IBM T. J. Watson Research Center, Yorktown Heights, USA

Simulating a $100 Million Data Center on a $100,000 Cluster
David Meisner and Thomas F. Wenisch (University of Michigan)

EuroCloud: Energy-conscious 3D Server-on-Chip for Green Cloud Services
Emre Özer α, Sachin Idgunji α, Krisztián Flautner α, Ali Saidi α, Yiannakis Sazeides β, 
Bushra Ahsan β, Nikolas Ladas β, Chrysostomos Nicopoulos β, Isidoros Sideris β, Babak 
Falsafi γ, Almutaz Adileh γ, Michael Ferdman γ, Pejman Lotfi-Kamran γ, Mika 
Kuulusa δ, Pol Marchal ε, Nikolas Minas ε
α ARM, β University of Cyprus, γ EPFL, δ Nokia and ε IMEC

Data Dwarfs: Motivating a Coverage Set for Future Large Data Center Workloads
Mehul Shah, Parthasarathy Ranganathan, Jichuan Chang, Niraj Tolia; HP Labs, Palo Alto, CA
David Roberts, Trevor Mudge; University of Michigan, Ann Arbor, MI

(For reference, here is the call for posters, which has concluded)
Ben Sigelman,
Mar 21, 2010, 8:37 PM
Ben Sigelman,
Jun 18, 2010, 10:22 PM