(High-End) Computing Systems Group

(High-End) Computing Systems Group

Department of Computer Science and EngineeringThe Ohio State University

• Computation spread over hundreds and thousands of processors

• Provides far-reaching benefits to society: – new drugs, – safer and fuel-efficient vehicles, – environmental modeling – weather and climate prediction – scientific discoveries in a broad range of disciplines

• Essential to commercial domains as well – Web servers, large databases, search engines

High-End Computing and Its Benefits

• Cover the entire range, from the Applications level, through the Systems Software level, to the Networking and Communications level, in developing high-end systems and enabling their use in application areas:

– Data-Intensive Computing– Data-mining and Database Systems– High-Performance Scientific Computing– Middleware and Compilers– Network-Based Computing

• Carry out research in an integrated manner (systems, networking, and applications)

Our Vision

• The cost/performance ratio of computing, networking, and storage components has improved dramatically, making the aggregate available “raw” computing power very high

• However, the fraction of the potential power that is actually utilized is getting smaller every year

• Two broad challenges:– Make the large aggregate power of distributed

computing resources available in a transparent manner to high-end applications

– Create high-level approaches to ease the development of high-end applications

The Research Challenges

Par/Dis Applications

Systems Software

Networking & Storage

Middleware & CompilersScheduling, DependabilityAnd Security

Bio-InformaticsDatabases and DataminingScientific Computing

Agrawal, Qin,SadayappanSaltz,

FerhatosmanogluParthasarathy

Communication ProtocolsActive Network Interfaces,Data Access

Research Areas Covered

Panda, LauriaZhang

• Gagan Agrawal – Grid Computing, Middleware, Data Mining, Biological Data Integration

• Hakan Ferhatosmanoglu – Databases

• Mario Lauria – I/O and Communication, Computational Biology

• D. K. Panda – Architecture, Communication, & Networking

• Srini Parthasarathy – Data-intensive computing & Data Mining

• Feng Qin– Operating Systems, Security and Dependability

• P. Sadayappan – Performance Optimization, Compilers, & Scheduling

• Joel Saltz – High Performance Computing Software & Bioinformatics

• Xiaodong Zhang – Distributed Systems, Memory Systems (also Networking)

Faculty Involved

Faculty, Students, Funding, and Accomplishments

• Over 60 graduate students involved in research in the Systems area – More than 45 funded as RAs

• Two post-docs• Research expenditure for FY ’05 ~ $2.5M• Several large-scale grants

– Four NSF medium-sized ITR (Saday, Joel(2), Srini)– DOE SciDAC (DK (2))– NSF RI Award (DK and the group) – COE:P award (Agrawal and others)

• Several CAREER Award Winners– Srini, Hakan (NSF and DOE)– Gagan (NSF)– DK (NSF)

• Six Best Paper Awards in major conferences during the last four years

• First employment of graduated students– Arizona State University, Louisiana State University,

IBM TJ Watson, IBM Research, Argonne National Lab, SGI, Compaq/Tandem, Pacific Northwest National Lab, Fore Systems, Microsoft, Lucent, Citrix, Dell, Yahoo, Amazon, Oracle, Ask.com, ….

• Several of the past and current students– OSU Graduate Fellowship– OSU Presidential Fellowship– IBM Co-operative Fellowship– CSE annual research award

Faculty, Students, Funding, and Accomplishments

• Investigators are strong players in many National-level Initiatives– NPACI– DOE Programming Models – HUBS/DARPA

• Collaborations with major industry and National Labs

• Open-source Developments and Distribution– Datacutter; also part of NPACkage– MVAPICH (MPI over InfiniBand)

• more than 400 organizations world-wide– TCE (Tensor Contraction Engine) for Computational

Chemistry

Participation in National-Level Initiatives and Open-Source

Developments

Projects: Gagan Agrawal

Overall Agenda: Efficient Processing of Data arising from distributed sources

Research Components: Middleware for Streaming Data (GATES)

Investigate self-adaptation, process migration, fault-tolerance .. Middleware for Data Analysis in Clusters and Grids (FREERIDE and

FREERIDE-G) Investigate Parallelization from high-level API, Remote Data Access

Automatic Data Virtualization and Wrapper Generation Frameworks Management of large scale data, especially from Sensors and Scientific

Experiments Biological Data Integration Others: Parallel Compilation, Query Optimization (XQuery), Data

Mining and OLAP algorithms

• Overall: Data Management in Modern Applications– Problem: Massive amount of Multi-dimensional data– Goal: Scalable systems for Efficient queries

• Multimedia Data Management• Image, audio, video, document databases (DBs)• Spatial DBs: GIS • Time-series DBs: stock market

• Bioinformatics & Biomedical Data Management– Genome data: DNA, proteins, gene expression data– Structural analysis of bio-chemical data

• Stream and Sensor Data Management– Telecommunications and internet data management– sensor networks: geo-sensors, bio-sensors

• Parallel I/O

Projects: Hakan Ferhatosmanoglu

Projects: D. K. Panda

• System Software/Middleware– High Performance MPI on InfiniBand Cluster– Clustered Storage and Parallel File Systems– Solaris NFS over RDMA – iWARP and its Benefits to High Performance Computing– Efficient Shared Memory on High-Speed Interconnects– High Performance Computing with Virtual Machines (Xen-IB)– Design of Scalable Data-Centers with InfiniBand

• Networking and Communication Support– High Performance Networking for TCP-based Applications – NIC-level Support for Collective Communication and

Synchronization– NIC-level Support for Quality of Service (QoS)– Micro-Benchmarks and Performance Comparison of High-Speed

Interconnects

• More details on http://nowlab.cse.ohio-state.edu/ Projects

Projects: Feng Qin• Agenda: Dependability and Security of Computer

Systems• Online Techniques for System Dependability and

Security– Failure recovery for high-end parallel and distributed

systems– Online bug diagnosis, i.e., identifying root causes of

software bugs– Dynamic system security enhancement based on runtime

information

• Software Debugging– Bug detection (esp. concurrency bugs) in multi-core, parallel

and distributed systems– Automated software bugs localization/isolation (e.g. across

different versions)

Projects: P. (Saday) Sadayappan

• Systems Support for High-Level Parallel Programming: Goal is to

enable higher level programming than message passing (MPI),

without sacrificing performance

– Automatic synthesis of high-performance parallel programs

for a class of quantum chemistry computations

– Compiler optimizations for locality enhancement and

communication minimization.

– Parallel Global-Address-Space programming with MATLAB

– Scheduling and load balancing

– Performance Optimization for multi-core architectures

– CIS 888.11K (every quarter)

• Agenda: Data Mining and Parallel/Distributed Systems.• Systems Support for Data Mining Applications

– Resource and Location Aware Data Management and Mining for Dynamic (potentially streaming) & Distributed Datasets.

– Distributed Shared State for Interactive Applications

• Fundamental Algorithms and Techniques– Incremental Techniques for Mining Streaming Datasets– Parallel and Distributed Data Mining Algorithms

• Applications Research– Intrusion Detection– Scientific & Biomedical Data Mining– Web/Text Mining

Projects: Srinivasan Parthasarathy

• Dynamic Data Driven Applications Systems– Data-intensive and Grid Computing tools and frameworks targeting

data/compute intensive applications– Runtime and compiler support– Component frameworks for combined task and data parallelism in

heterogeneous environments – Service-oriented Architectures for Grid-enabled data-intensive computing

• Scheduling Services in the Grid• Grid Generalized Reduction• Active Semantic Data Caching• Data Cluster/Decluster/Range Query Services

• Application Areas include– Earth Systems Sciences: Instrumented oil field simulations, seismic data

analysis.– Medical Imaging: Texture analysis, segmentation, registration of ensembles

of multi-modal, multi-resolution, time-dependent imagery.– Pathology Informatics: Visualization and exploration of digitized pathology

slides– Bioinformatics: Querying and analysis of large databases of gene and protein

sequence data. – Medical Informatics: Ad-hoc, federated data warehouses.

Projects: Joel Saltz (Bioinformatics and CSE)

Putting Disk Layout Information on the OS map

building a Disk-Seen system to exploit Dual LOcality (DULO): temporal locality and disk spatial locality.

DULO-caching and DULO-prefetching.

Disk Energy Saving

caching and prefetching in flash drive.

Multi-level disk caching and prefetching.

Cooperative I/O buffer caching in large clusters.

DNS caching consistency

Xiaodong Zhang: Data Xiaodong Zhang: Data Access in Core and Access in Core and

Distributed SystemsDistributed Systems

Wyckoff: High-performance Storage

• Investigate:– Object-based storage devices used in parallel file

systems• Goal:

– Enable greater scalability and higher performance of large storage systems by interacting with disks at a higher semantic level.

• NSF-funded project starting Sep 06 for 3 years, looking for student(s)

• Involves aspects of systems, storage, protocols and networking

• Helpful skills:– C programming– Unix/Linux systems– Parallel computing– File systems

• Contact: [email protected]

High-End Computing and Networking Research Testbed for Next

Generation Data Driven, Interactive Applications

PIs: D. K. Panda, G. Agrawal, P. Sadayappan, J. Saltz and H.-W. Shen

Other Investigators: S. Ahalt, U. Catalyurek, H. Ferhatosmanoglu, H.-W. Jin, T. Kurc, M. Lauria, D. Lee, R.

Machiraju, S. Parthasarathy, P. Sinha, D. Stredney, A. E. Stutz, and P. Wyckoff

Dept. of Computer Science and Engineering, Dept. of Biomedical Informatics, and Ohio Supercomputer Center

The Ohio State University

Funded by NSF Research Infrastructure (RI) ProgramAward: $3.1M = $1.53M (from NSF) + $1.48M (Cost-Sharing from OBR and

OSU)

Contact: [email protected]

Our Vision of the Next Generation Architecture

Client 1 Client 2 Client N

ComputingEnvironme

nt- Basic Processing- Post-processing

DataRepositor

y10-100

TeraBytes

- Pre-processing

Wide Area

Network(WAN)

Wired or Wireless Network

MemoryCluster

LAN/SAN

Interactive

Collaborative

End-to-end QoS

Experimental Testbed Installed

CSE

70-node Memory Cluster with512 GBytes memory, 24 TB disk

GigE, and InfiniBand SDRUpgrade (Yr4)

BMIMass Storage System

500 TBytes(Existing)

OSC

64-node Compute Cluster with InfiniBand DDR and 4TB disk

10 GigE on some (to be added)Upgrade (Yr4)

Graphics Adaptersand Haptic Devices

20 Wireless Clients & 3 Access PointsUpgrade (Yr4)

Video wall

2x10=20 GigE 40 GigE (Yr4)

2x10=20 GigE 40 GigE (Yr4)

10.0 GigEswitch

10.0 GigEswitch

10.0 GigEswitch

Collaboration among the Components and

Investigators

Networking, Communication,

QoS, and I/O

Programming Systems and Scheduling

Data Intensive Algorithms

Data Intensive Applications

Panda, Jin, Lee, Lauria, Sinha, Wyckoff, and Kurc

Saltz, Agrawal, Sadayappan, Kurc, Catalyurek, Ahalt, and Hakan

Shen, Agrawal, Machiraju, and Parthasarathy,

Saltz, Stredney, SadayappanMachiraju, Parthasarathy, Catalyurek, and Other

OSU collaborators

Wright Center for Innovation (WCI)

• A new funding to install a larger cluster with 64 nodes with dual dual-core processors (up to 256 processors)

• Storage nodes with 40 TBytes of space• Connected with InfiniBand DDR• Focuses on Advanced Data

Management

High Perf. Comp.

Architecture

Operating Systems

Databases/DataMining

Languages/Compilers

CSE 621

CSE 775

CSE 760

CSE 671

CSE755

CSE 721

CSE875

CSE 762

CSE772

CSE756

• 621, 756 are only offered in Autumn• 721 is only offered in Winter• 875 is only offered in Spring (775 in Au/Sp)

CSE770

Relevant Courses

• 788.xxx: 3 credit, letter-graded, once in two years

• 888.xxx: S/U graded, every quarter• Agrawal: 788.11I, 888.11I• Ferhatosmanoglu: 788.02H, 888.02H• Lauria: 788.08R, 888.08R• Panda: 788.08P, 888.08P• Parthasarathy: 788.02G, 888.02J• Sadayappan: 788.11J, 888.11J,

888.11K

Specialty Courses

• Addressing cutting-edge research challenges with focus on multi-disciplinary applications

• Significant research funding from federal, industrial and state sources

• Synergistic group with significant growth in the last few years

Summary