CAECW-10 Workshop Program

 

 

9am – 10am: Keynote

 

The Impact of Virtualization on Commercial Workloads

Mendel Rosenblum (Stanford University)

 

This talk describes how virtualization is changing the way computing is done in the industry today and how it is causing users to rethink how they view hardware, operating systems, and application programs.  The talk will describe this new view on computing and the benefits driving users to adopt it. The changing roles for hardware and operating systems will be discussed along with what changes will be needed to efficiently and simply support this new computing model.

 

Bio: Mendel Rosenblum is an Associate Professor in the Computer Science Department at Stanford University. Professor Rosenblum research interests include system software, distributed systems, and computer architecture. He has published research in the area of disk storage management, computer simulation techniques, scalable operating system structure, virtualization computer security, and mobility. He is also a co-founder of VMware Inc. As the Chief Scientist of VMware he helped design and build the industry-leading virtualization technology for commodity computing platforms. He is the recipient of the 2002 ACM/SIGOPS Mark Weiser Award for creativity and innovation in operating systems research. He received a BA in Math from the University of Virginia (1984) and a MS (1989) and PhD (1992) in Computer Science from the University of California at Berkeley

 

10.00am – 10.30am: Coffee break

 

10:30am – 12:30am: Session 1: Java Workloads

 

O Java, Java! Wherefore Are Thou Java? (Invited paper)

K. S. McKinley (The University of Texas at Austin) and S. M. Blackburn (Australian National University)

 

The experimental systems community relies on benchmarks to evaluate proposed innovations in applications, compilers, operating systems, and architecture.  Therefore, the choice of benchmarks and the programming language in which they are written is a gating function for innovation in our field.  While the applications community is increasingly embracing managed languages, such as Java and C\#, due to their software engineering benefits, systems researchers have not. For the most part, systems researchers use C or C++.  To evaluate managed languages, researchers will need new evaluation methodologies. We demonstrate and recommend how to perform meaningful experiments as compared to those in use for C or C++.  Because many applications now and in the future will be in managed languages, it may behoove some systems researchers to eat their own dog food, by building systems in managed languages.

 

Behavior Characterization and Performance Study on Compacting Garbage Collectors with Apache Harmony

C. Lai, I. T. Volosyuk, X.-F. Li (Intel)

 

A Study of Instruction Cache Performance and the Potential for Instruction Prefetching in J2EE Server Applications

P. Nagpurkar (University of California, Santa Barbara), H. W. Cain, M. Serrano, J.-D. Choi (IBM), C. Krintz (University of California, Santa Barbara)

 

CMP/CMT Scaling of SPECjbb2005 on UltraSPARC T1

D. Kaseridis and L. K. John (The University of Texas at Austin)

 

12:30am – 2:00pm: Lunch

 

2:00pm – 3:00pm: Session 2: System Workloads

 

Simulating Complex Enterprise Workloads using Utilization Traces (Invited paper)

P. Ranganathan and P. Leech (HP)

 

Performance Analysis of Snoop Filter Replacement Policies in Multi-Bus Commercial Server Platforms

S. Chinthamani, M. Mandviwalla (Intel)

 

3.00pm – 3.30pm: Coffee break

 

3:30pm – 5:00pm: Session 3: Emerging Workloads

 

Optimizing Data Mining Workloads using Hardware Accelerators (Invited paper)

A. Choudhary, R. Narayan, B. Ozisikyilmaz, G. Memik (Northwestern University), J. Zambreno (Iowa State University), J. Pisharath (Intel)

 

Data mining is the process of finding useful and actionable patterns in large data sets. Datamining algorithms have become vital to researchers in science, engineering, medicine, business, search and security domain. In recent years, there has been a tremendous increase in both the size of the data being collected that must be analyzed. Processors and systems, on the other hand, have been designed and optimized for simulation, scientific, media, and database workloads. In our previous work in the development of Minebench, a comprehensive data mining benchmark, we have demonstrated that computational characteristics as well as data access requirements for datamining workloads is quite different than those of these other workloads. In this paper, we present a brief overview of the current approaches and challenges faced in system design. Then we present initial designs and results for accelerating data mining algorithms using programmable hardware. Initial results show that tremendous performance gains can be obtain by accelerating these workloads over using traditional systems.

 

Performance Characterization of Parallel Replay Detection Video Mining Workload

E. Li, W. Hu, N. Di and Y. Zhang (Intel)

 

Parallellization and Performance Characterization of a New Multi-Document Summarization Method

J. Shan, Q. Diao, Y. Chen and Y. Zhang (Intel)

 

5:00pm – 5:15pm: Wrap Up