sun initiatives in grid and cloud computing: configuring ...-+sub+initiative.pdf · sun...
TRANSCRIPT
Stephen Perrenod, Ph.D.Director, HPC Business Development, APACSun Microsystems2008. 09. 15
INNOVATION MATTERS
Sun Initiatives in Grid and Cloud Computing: Configuring the Cloud
Grid Asia 2008
Agenda• 1982• Grid Engine
> Compute, data, visualization grids• Network.com Cloud• SOA• Grid-in-a-Box• Project Caroline• Project Hydrazine• Ranger
> Largest open supercomputer in world
Grid – Cloud - SOA
1982
1982
• Sun founded ”Network is the Computer”• Supercomputing service bureaus popular
> 56 kbit link
Our Vision• The Network is the Computer• Everyone and everything participates on the network
Int ernet User s
2000 2007 20102005
1.5 Billion
1995
Grid Engine
The Sun HPC Stack Core Solaris and Linux products from Sun
Sun C
RS, S
uppo
rt, A
rchit
ectu
ral, P
rofes
siona
l Ser
vices
Developer ToolsDeveloper Tools
Operating SystemOperating System
NodeNodeProcessor
ManagementManagementWorkload ManagementCluster Management
InterconnectInterconnect
Sun XVM Ops Centre Software
Sun Grid Engine Software
Ethernet, Myrinet, InfiniBand
Free
Open &SupportedVersions
64 Bit
Open
Solaris isOpen &Free
Open
Sun Studio
Sun HPC Cluster Tools
Tokyo Institute of Technology TSUBAME on Grid Engine> Accounting
> Grid Engine provides detailed accounting information> Tight integration with rsh and ssh
> Lots of machines> Grid Engine is able to scale to their needs
> Lots of MPI> Grid Engine's configurability makes integration possible
> Lots of applications> Grid Engine is zero-touch – most apps won't notice
> Infiniband support> Absolutely!
Sun Grid Engine grid #2
Sun Grid Engine Multi-Clustering
Sun Grid Engine grid #1
Service Domain Manager
• Grids are monitored by Service Level Objectives• Policies control relative grid priorities
Sun Grid Engine grid #1
Sun Grid Engine grid #2
Spare Pool
Service Domain Manager
I need resources
I have 2 free
Sun Grid Engine Multi-Clustering
Sun / APSTC Bio-Cluster Grid• http://apstc.sun.com.sg - BioBox initiative
> Bio-Cluster Grid> Over 20 of most popular bioinfo apps ready to run on
Linux or Solaris clusters> using Sun Grid Engine for workload management
• Seamless, simplified and GUI access to high throughput computing resources on the grid
• Access to dispersed data resourcesJoint Collaboration between NUS and APSTC
The Architecture
Grid Engine Portal
Sun Grid Engine
Dispatch
Results
Jobs
Cluster of Execution Hosts
Bio-ClusterGrid: Benefits • Fast setup (3 hours)
> Installation of the 28 Bioinformatics applications> Activation of Sun Grid Engine > Installation and Activation of Grid Engine Portal/JES Portal
Server• Compared to Manual Installation (9 days)
> Downloading, compiling and installation of 28 bioinformatics applications (1 week)
> Installation and Configuration of Sun Grid Engine and Portal (2 days)
• Huge saving in time for administrators> No need to deal with the complexity of compilation :
compilation errors, checking for installing dependencies.
Sun / APSTC Bio-Cluster Grid Sites• APAC Bio-ClusterGrid sites:
> Genome Institute of Singapore > BIOTEC - Thailand > UKM & USM Grid Testbed - Malaysia > School of Biological Sciences, NTU - Singapore > School of Chemical and Life Sciences - Singapore Polytechnic > Centre for DNA Fingerprinting & Diagnostics - India > National Cheng Kung University - Taiwan > International Islamic University - Malaysia > University of Delhi South Campus - India
Data and Visualization Grids
Sun's HPC Three Tier Storage Architecture for Information Life-cycle Management
Compute ClusterIntermediate Data
& Cache
Primary Storage
Backup
High speed, High I/O Cluster
facing
Medium speed, enterprise class
Lower speedhigh capacity
archival facility
Sun Visualization Grid System Sun Shared Visualization Sun Scalable Visualization
Supports combination of multiple graphics devices to drive:> Higher Performance> Higher Image Quality> Higher Resolution
NetworkGraphics
Graphics
• Scalable• Sharable• Integratable• Secure• Virtualizing Visualization
Network.com
Cloud Computingfrom Sun
Innovative Business Models: The Sun Grid – now in 25 countries
$$11$1/CPU-hr$1/CPU-hr
Computing Becomes an Operational Expense
• No hidden fees• No minimums• No barrier to exit• Immediate access• Simple licensing ISVs• www.network.com
Network.com Application CatalogEnabling On-Demand Delivery of HPC Applications
• HMMER • MPQC • NAMD • PETSc• POP • Q-Chem • QCM • Rational Numbers Assign • Rational Numbers Partition • Rational Numbers FragSearch • Readseq • T-Coffee
• Life Sciences Apps:• Amber PMEMD-parallel• ARMS • ASV • BLAST • ClustalW • CSR• DOG• eHiTS • EMBOSS • FASTA • fastDNAml • GROMACS
Network.com users in Life Sciences
• SimBioSys eHiTS> By using the computationally intensive eHiTS program on
the Sun Grid Compute Utility, scientists can leverage additional compute capacity to speed time to results for molecular docking, and accelerate the pace of innovation and discovery.
• Applera / Applied Biosystems> Through Sun Grid Compute Utility, AB was able to perform
the compute-intensive data research to develop millions of new genomic assays in a matter of days rather than months. In addition, because the company only had to purchase the number of hours required, at a rate of $1 per CPU hour, it avoided an investment in infrastructure that would have cost the company hundreds of thousands of dollars.
SOA
Customer Needs Solution
• Vendor Mergers & Acquisitions • SOA-based Architectures Leverage• Legacy Systems• Image Archiving
• Digital Information Flow • Management Of Escalating Data • Regulatory Privacy& Access Control
• Health Information Networks • Federated IdM and Compliance
• Consumer-directed Healthcare• Long Term Care Management For Higher• Quality And Lower Cost
• Genomic Profiling And Pharma Data• Guidance Based On Best Practices• Data Privacy
• Personal Health Records• Disease Management Platforms
• Electronic Medical Records • Identity Management & Security
• Business Integration & Composite • Applications
Digitization
Consumerism(Patient-centric)
Evidence- based Medicine
Consolidation
Healthcare Industry Trends
Integrating Medical Images with EMR Reduces Access Time
Single Patient View of Medical Record Includes Images
ANYTIME, ANYWHERE
ACCESS TO A PATIENT’S
IMAGES AND TEXT HISTORY
Healthcare “Grid” Customers• Sun Integration Solutions in Action:• UK NHS National Programme for IT - Sun Java Integration Suite used to help
deliver and manage a national patient record database and transactional messaging service
• Sweden Capio AB - data flow from Oracle to local Patient Admin systems, extracted using Sun (Java CAPS) SeeBeyond eGate integrator and then fed to ERP
• Austria Oberosterreichische Gesundheits-und-Spitals-AG - large hospital operator uses Sun software to manage info flow from 400 interfaces across 50 heterogeneous IT systems
• UK Salford Royal NHS Hospital Trust - Sun software to manage integration of info from patient admin, pathology, radiology and other systems with central electronic patient record, all on one screen
• Luxembourg Le Centre Hospitalier de Luxembourg - uses SeeBeyond (Java CAPS) to link SAP-based Hospital Information System with Radiology, Cardiology, Theater, Laboratory and PACS (Picture and Archiving System); also used to exchange info with other hospitals
• Sweden Kalmar County Council - one patient, one record with Cambio and Sun SeeBeyond (Java CAPS) eInsight Business Process Manager
Sun Integration Technology• Pragmatic approach to SOA ("An architecture where
services are defined and orchestrated using open standards, allowing for a pluggable, agile, heterogeneous service infrastructure")
• Application-to-application Integration - Java CAPS / SeeBeyond (unified suite to develop, deploy, manage, and monitor a SOA)
• Information Lifecycle Management – StorageTek storage and SAM-FS archival file system
• Virtualization (Containers, Domains)• Security - Thin client Sun Ray• Broad range of OS platforms: Solaris, Linux,
Windows• Broad range of hardware: Sparc and CMT, x86 (Intel,
AMD)
Grid-in-a-Box
Sun Modular Datacenter S20 / D20
Standard shipping container packaged with eight standard racksIntegrated, high‐efficiency power and coolingSupports a wide range of compute, storage and network infrastructure – build once, deploy anywhere when fully configured Top 250 with over 250 servers configurable
Stanford Linear Accelerator• Stanford Linear Accelerator Center (SLAC) High Performance Computing Node• Supports Particle Physics Research
> Babar experiment (B-Mesons)> The goal of the experiment is to study the
violation of charge and parity (CP) symmetry in the decays of B mesons. This violation manifests itself as different behaviour between particles and anti-particles and is the first step to explain the absence of anti-particles in everyday life.
Project Caroline
Project Caroline (projectcaroline.net)• Sun Labs project - open source research project• Developing a horizontally scalable platform for the
development and deployment of Internet services
• The platform comprises a programmatically configurable pool of virtualized compute, storage, and networking resources
• Project Caroline helps software providers develop services rapidly, update in-production services frequently, and automatically flex their use of platform resources to match changing runtime demands
• Caroline Platform API 10.4.2 Released August, 2008
Project Caroline http://blogs.sun.com/zippel/entry/project_caroline_video
• resources: computation, storage, network• process - java perl python (can put ruby or php on top of those)• file systems and DBs, use ZFS can propagate many ind. file systems (via NFS,
WebDAV), postgres DB (mysql later), private & internet networks; Load balancer, NAT, VPN; DNS mapping
• resource creation, modify, destruction etc. and can monitor• create ip addresses for services, DBs, (clients), and address for domain name• process creation - includes firewall rules, IPs, file systems• can clone for debug, test• layer 4 load balancing• can bind directly, static, dynamic net for outbound ....• public DNS also private namespace for own account• Demo of building and taking down
Data Center Virtualization ApproachesComputation
Bare processors
HW VM's, e.g, Xen,LDOM, & VMWare
Language VM'se.g., Java, Python, etc.
Domain specificlanguagesFortress
OS, e.g. Solaris, LinuxZones
Incr
easin
g vir
tuali
zatio
n
StorageCommunications
KnowledgeSpaces
Object StoreDHT, MySQL
pNFS, distr. FS, GFS
File SystemsZFS
Volumes
Disk BlocksSATA, SCSIBare Disks
MapReduceMaster Worker
MPICrossbow
TCP/IP network
Ethernet, WiFiInfiniband
Bare Cables
Project Caroline Google
Bare Metal
Hypervisor Based Virtualization
Project Hydrazine
Vision and Objectives• Vision
> Hydrazine enables the rapid mashing of services creating the “rocket fuel” that will power millions of Java and JavaFX applications delivering personalized, contextual based services across 4 screens
• Objectives> Provide a complete development environment enabling the rapid creation and
deployment of rich cloud applications> Simplify the discovery and utilization of key service enablers facilitating the
creation of compelling blended, personalized, and contextual services> Provide means for developers to more easily monetize their services> Drive developer adoption to support the creation of back end services for
JavaFX and Java applicationsProject Hydrazine AWS is “the launch pad, not just for the next million Facebook apps, but
also for personal live TV channels, virtual desktops, pay-by-the-mile auto insurance, and no doubt plenty of things no one has thought of yet” (Wired Magazine, Apr 08)
Project Hydrazine Hosted, standards-based platform enabling developers to easily
discover, blend, deploy, and monetize services
Service Repository Deployment Platform
3rd Party Services
App App App
GlassFish
Project
Hydrazine
Network.com
Open ESB
mySQL
Service EnablersComms Suite
3rd Parties Service SP Services
...
Storage
Database
Core ServicesIdentityProfileSession...
App EnablersmyMedia
Address Book AggContextual mapping
...
Ranger at TACC
• The world’s largest general purpose compute cluster in production (4 Feb. 2008) based on Sun Constellation System
• National Science Foundation funded TeraGrid center – more capacity than all others combined
• Now 580 Tflops> 82 Sun ultra-dense blade platforms> 2 Sun ultra-dense switches> 72 Sun X4500 storage servers
• Sun is the sole HW supplier• Opteron quad-core based• Other customers in the pipeline, from small to large
The First Sun Constellation System in Production – Feb. 2008
Courtesy of Indiana University
Sun Constellation System Open Petascale Architecture
OtherCablingInfrastructure
OtherComputeClusters
ReducedCabling
Racks
CablingCoreSwitches
Leaf Switches
• 1 switching element 300:1 reduction• 1152 cables 6:1 reduction• 74 racks 20% smaller footprint
Radical Simplicity, Faster time to Deployment
Constellation SystemOpen Super Computer
RacksSunConstellationSystemCluster
Cabling
• 300 switching elements• 6912 cables• 92 racks
Alternative Open Standards Fabric
Why Does Sun Care about HPC and Grid?• “Redshift” sectors include HPC – the Universe is Expanding
> Grid, Cluster computing and Networking is the driver• It's a growing market, outpacing growth of many others
> $20B opportunity in 2008• Success in HPC is a precursor to success in Enterprise markets
HPC is under served by Moore's Law