September 2012 - talks - lists.cs.princeton.edu

Colloquium Speaker Hyesoon Kim Oct 3
by Nicole E. Wagenblast 01 Oct '12

01 Oct '12

When GPUs meet CPUs: opportunities, challenges and solutions in heterogeneous architectures Hyesoon Kim , Georgia Institute of Technology Wednesday, October 3, 2012- 4:30pm Computer Science Small Auditorium, 105 The last decade has seen a paradigm shift in the architecture of computing platforms: Uni-processors have given way to multi-core (many-core) processors, and now the industry is moving toward heterogeneous architectures that combine CPUs and GPUs on the same chip, as we can see from Intel, AMD, NVIDIA, etc. Heterogeneous architectures are especially attractive as they can provide high performance and energy-efficiency for both general-purpose applications as well as high throughput applications. However, these architectures introduce several new challenges: including programming, determining power and performance trade-offs and developing hardware solutions that exploit the underlying heterogeneity. In this talk I will present some of our recent work that reduces the software effort required in programming such architectures, and provides hints to estimate the performance and power behavior of CPUs , GPUs or CPUs+GPUs. I will also discuss architecture solutions that improve overall system performance by taking into account the difference in characteristics of CPU and GPU applications, and optimizing the cache partitioning, prefetching, and DRAM scheduling to best suit the workload needs. Hyesoon Kim is an Assistant professor in the School of Computer Science at Georgia Institute of Technology. Her research interests include high-performance energy-efficient heterogeneous architectures, programmer-compiler-microarchitecture interaction and developing tools to help parallel programming. She received a BA in mechanical engineering from Korea Advanced Institute of Science and Technology (KAIST), an MS in mechanical engineering from Seoul National University, and an MS and a Ph.D in computer engineering at The University of Texas at Austin. She is a recipient of the NSF career award in 2011.

1 1

Canturk Isci seminar. Today at 12:30pm in CS302.
by Margaret Martonosi 25 Sep '12

25 Sep '12

Reminder: Dr. Canturk Isci from IBM Research will be speaking here today (Tuesday Sept 25) at 12:30pm in CS 302. Virtualization Management and Data Center Monitoring for Energy-efficient Clouds Canturk Isci IBM Research Abstract: Energy and thermal management in the cloud has become a primary concern as the global footprint of large-scale compute systems reaches significant levels, and continues to increase with impressive pace, following the ever-increasing compute demand and data explosion. Various new control knobs across systems, middleware, virtualization and facilities levels promote many interesting optimization opportunities for improving energy efficiency. However, these are coupled with significant management and monitoring challenges for effectively exploiting the present opportunities. In this talk we first look at some of the potential opportunities and emerging challenges for improving the energy-efficiency in clouds, and highlight our research at virtualization, distributed systems and data center levels targeting these opportunities and challenges. We describe an effective distributed resource management approach based on a novel, black-box virtual machine (VM) demand estimation technique, which serves as a key enabler for improving the energy efficiency of virtualized systems via consolidation-driven, dynamic VM placement. We demonstrate that this demand estimation technique is highly-general across virtualization technologies and can very accurately track actual VM demands even under extreme consolidation and resource constraints. With multiple real-system prototypes, we present the dramatic improvements in the performance and agility of energy-aware virtualization management when guided by demand estimation. Next, we briefly look at the increasing cost and complexity of monitoring in the cloud and describe a new approach based on mobile, autonomous robots for reliable, fast and cost-effective monitoring and management in data centers. We demonstrate the operation principles of actual robot prototypes that are deployed in various live data centers, and discuss their applications to predictive and diagnostic analytics for energy, thermal and asset management. Bio: Canturk Isci is a Research Staff Member in IBM TJ Watson Research Center. His research interests include virtualization, data center energy and thermal management, microarchitectural and system-level techniques for energy-efficient and adaptive computing. Prior to joining IBM Research, Canturk was a Senior Member of Technical Staff at VMware, where we he worked on distributed resource and power management, performance and scalability of virtualized systems. He is the recipient of a best paper award in ICAC 2011, best research poster in VMworld 2008 and academic fellowships from British Council, Princeton and Bilkent University. He serves as the industry chair in IEEE Computer Society, Special Technical Community on Sustainable Computing. Canturk has a B.S. in Electrical Engineering from Bilkent University, an M.Sc. with Distinction in VLSI System Design from University of Westminster, and a Ph.D. in Electrical Engineering from Princeton University. Margaret Martonosi Professor, Dept. of Computer Science Princeton University http://www.princeton.edu/~mrm mrm(a)princeton.edu

1 0

12:30-1:20pm Mon Sep 24 on network operating system for the cloud
by Jennifer Rexford 21 Sep '12

21 Sep '12

Speaker: Dr. Anil Madhavapeddy, University of Cambridge Title: A Library-based Network Operating System for the Cloud Date/time: 12:30-1:20pm Monday September 24 (lunch starts at noon) Location: 302 in the Computer Science building Abstract: The cloud has so far mostly been used to consolidate existing operating systems and manage them more conveniently. The interface between guest kernels, applications and VMs are all managed very differently, leading to serious inefficiencies, unreliability and insecurity. Our new Mirage OS revisits the library OS, and finds it a practical approach to improve this situation. Applications are written in a high-level functional programming language (OCaml) and compiled directly into microkernels that run on the Xen hypervisor. By treating the hypervisor as a stable hardware platform, we can focus on high-performance protocol implementations without worrying about having to support the thousands of device drivers found in a traditional OS. Mirage includes clean-slate functional implementations of protocols ranging from TCP/IP, DNS, SSH, Openflow (switch/controller), HTTP, XMPP and Xen inter-VM transports. In this talk, I will describe the basics of Mirage, and demonstrate the performance and resource tradeoffs with our approach. Some of the new applications we are building using Mirage include the next-generation Xen Cloud Platform (a widely deployed open-source Xen distribution), Openflow-integrated applications, and a novel approach to managing distributed devices on the Internet.

1 0

W Kim FPO Wednesday
by Melissa M. Lawson 17 Sep '12

17 Sep '12

As previously announced, Won ho Kim will present his FPO on Wednesday September 19 at 2PM in Room 402. His committee will now be: Vivek Pai, David Walker, and Brian Kernighan. Everyone is invited to attend his talk.

1 0

P Prabhu preFPO
by Melissa M. Lawson 12 Sep '12

12 Sep '12

Prakash Prabhu will present his preFPO on Tuesday September 18 at 11AM in Room 402. The members of his committee are: David August, advisor; David Walker and JP Singh, readers; Margaret Martonosi and Kai Li, nonreaders. Everyone is invited to attend his talk. His abstract follows below. -------------------- Semantic Language Extensions for Implicit Parallel Programming Several emerging ﬁelds of science of engineering are increasingly characterized by computationally intensive programs that generate exponential amounts of data. Without parallelization, such programs do not benefit from the increasing core counts available in today’s chip multiprocessors. However, it is widely perceived that writing correct and well performing parallel programs is an extremely hard problem. In order to understand the challenges faced by scientific programmers in effectively leveraging parallel computation, this dissertation presents an in-depth field study of the practice of computational science. Based on the results of the ﬁeld study, two new implicit parallel programming (IPP) solutions are proposed. IPP solutions offer the best of both automatic and explicit parallelization. With IPP, artificial constraints imposed by sequential models for automatic parallelization are overcome by use of semantic programming extensions that do not sacriﬁce the ease of sequential programming. These extensions are exploited by transformation tools to synthesize a correct parallel program without requiring enormous manual programming effort associated with explicit parallelization. The first IPP solution, Commutative Set, generalizes existing notions of semantic commutativity. It allows a programmer to expressively relax execution orders prohibited under a sequential programming model and enables multiple parallelism forms without additional parallelism constructs. The second IPP solution, WeakC, provides language extensions to weaken semantics of sequential data structures, and dynamically optimizes a parallel configuration of these data structures via a combined compiler-runtime system. Both Commutative Set and WeakC have been implemented within the clang/LLVM compiler framework, and are evaluated on several real-world applications running on real hardware, including some that are actively used by several scientists in their day-to-day research. Based on experimental results, this dissertation demonstrates the effectiveness of the proposed techniques.

1 0

W Kim FPO
by Melissa M. Lawson 12 Sep '12

12 Sep '12

Wonho Kim will present his FPO on Wednesday September 19 at 2PM in Room 402. The members of his committee are: Vivek Pai, advisor; Michael Freedman and Larry Peterson, readers; Jennifer Rexford and David Walker, nonreaders. A copy of his dissertation "Understanding Resource Usage and Performance in Wide-Area Distributed Systems", is on display in Room 310. Everyone is invited to attend his talk.

1 0