Clouds for research computing
28
0
0
Full text
(2) UVIC HEP Research Computing. Outline • Research computing • Cloud computing • Clouds for research computing • Distributed research cloud for particle physics and astronomy • Outlook. Randall Sobie. IPP/University of Victoria. 2.
(3) UVIC HEP Research Computing. Research Computing Why is the universe not made of equal amounts matter and antimatter ?. We build instruments large detectors to record the collisions of matter and antimatter SLAC National Accelerator Lab. Record billions of particle collisions “Events”. Randall Sobie. IPP/University of Victoria. 3.
(4) UVIC HEP Research Computing. Research Computing Desktop computing 12+ processor cores 10+ TB storage 10G network cards Applications Interactive computing Visualization Remote sensor control. Randall Sobie. IPP/University of Victoria. 4.
(5) UVIC HEP Research Computing. Research Computing Cluster computing 300+ cores/rack 1000+ TB storage Interconnectivity Application High Throughput HTC Small scale parallel jobs (physical sciences). Randall Sobie. IPP/University of Victoria. 5.
(6) UVIC HEP Research Computing. Research Computing. Super Computers TOP500 100K cores High-speed interconnectivity. Applications Climate modeling Cosmology. Randall Sobie. IPP/University of Victoria. 6.
(7) UVIC HEP Research Computing. Research Computing. Computing Grids Distributed compute clusters 100K+ cores 1000TB ++ storage. Applications Very High Throughput Particle physics Earth sciences Health sciences. Randall Sobie. IPP/University of Victoria. 7.
(8) UVIC HEP Research Computing. Cloud computing. IaaS Infrastructure as a Service. IaaS Delivers computing infrastructure as a service Science clouds Commercial clouds (EC2) HTC platform. Randall Sobie. IPP/University of Victoria. 8.
(9) UVIC HEP Research Computing. Cloud computing. IaaS PaaS Infrastructure as a Service Platform as a Service. PaaS Delivers computing platform or software stack as a service Commercial clouds (EC2) eg Instrumentation control. Randall Sobie. IPP/University of Victoria. 9.
(10) UVIC HEP Research Computing. Cloud computing. IaaS PaaS Infrastructure as a Service Platform as aSaaS Service Software as a Service. SaaS Delivers applications or software stack as a service Commercial clouds (EC2) eg Mathematica or Matlab. Randall Sobie. IPP/University of Victoria. 10.
(11) UVIC HEP Research Computing. Why use clouds? The BaBar project stopped recording data in 2008. Complex code developed over 15 years Limited to specific operating systems and libraries Diminishing resources and few people. Virtualization is the only solution for preserving the software environment. !. Randall Sobie. IPP/University of Victoria. 11.
(12) UVIC HEP Research Computing. Data preservation Topical issue for many research fields Particle physics community has largely ignored this issue often because new facilities make the old data obsolete. Projects are now decades long and unlikely to be repeated for many years. We need to ensure the data is accessible for the long term. Challenge We need to preserve the software environment Randall Sobie. IPP/University of Victoria. 12.
(13) UVIC HEP Research Computing. Requirements. Sophisticated user communities Non-GUI users Batch computing. Complex software packages and demanding system requirements Specific OS system Specific application libraries and compilers. !. Medium-scale data sets (100s TBs) Data accessed (on-demand) from remote repositories. Randall Sobie. IPP/University of Victoria. 13.
(14) UVIC HEP Research Computing. Conceptual design of a distributed cloud Design goals: Leverage existing work in grid computing (authentication, data management, networks) Use existing research computing facilities and get access to new resources Boot user-customized VMs in a familiar batch computing environment Simplify systems configuration by removing the application dependence Use the network to move data to the clouds. Sky Computing or Grid of Clouds Randall Sobie. IPP/University of Victoria. 14.
(15) UVIC HEP Research Computing. Components Virtualization Clouds WS interface eg Nimbus, OpenStack, EC2. Application encapsulation Image replication eg Xen, KVM. I IaaS. Dynamic resources eg Condor, SGE. Cloud Scheduler. Managing multiple clouds eg Cloud Scheduler. Job Scheduler. Randall Sobie. IPP/University of Victoria. 15.
(16) UVIC HEP Research Computing. VM management : Repoman Repoman Is an image repository for storing standard or base VMs. It is also used by the users to store and manage their images. The user adds the http link to the VM into their job requirements file.. Randall Sobie. IPP/University of Victoria. 16.
(17) UVIC HEP Research Computing. CERN-VM Filesystem (CVMFS) CVMFS is a virtual software appliance designed to provide dynamic installation of application software CVMFS retrieves only the files required by the user Implemented as a FUSE module (File System in User Space) Advantages: • BaBar software suite is 30G • User analysis reduces size to 4 G • Boot/save times significantly decreased • Reduced load on network transfer of VM images. Randall Sobie. User requires an https/http connection to a read-only file system. IPP/University of Victoria. 17.
(18) UVIC HEP Research Computing. Authentication We use X509 certificates for authentication (except Amazon EC2) Used in particle physics (LHC/CERN) and also by Westgrid (Compute Canada) Certificates issued by Grid Canada X509 is an ITU Telecommunication standard for a public key infrastructure (PKI) for single sign-on We use it to X509 certificates for user job management, access to Repoman and access to the data storage. Randall Sobie. IPP/University of Victoria. 18.
(19) UVIC HEP Research Computing.
(20)
(21)
(22)
(23)
(24) . !. !. Randall Sobie. IPP/University of Victoria. 19.
(25) UVIC HEP Research Computing. IaaS cloud resources Access to few-100s cores on science clouds in Canada and the United States All sites (except EC2) on the research network UVIC-HEP Nimbus. NRC-Ottawa Nimbus. Integrating other FutureGrid sites in the distributed cloud. Amazon EC2. UVIC-Westgrid Nimbus FutureGrid Argonne NL Nimbus. Randall Sobie. IPP/University of Victoria. 20.
(26) UVIC HEP Research Computing. Access to 8 clouds (Nimbus, Eucalyptus, EC2) VMs retrieved from two repositories (except EC2) (implementing a VM cache) Data located at two sites. CS looks at the job queue and sends a request to the next available cloud to boot the User-VM Randall Sobie. IPP/University of Victoria. 21.
(27) UVIC HEP Research Computing. User view of the system is the same as a standard batch environment Job script contains a link to the user’s VM required for the job. Randall Sobie. IPP/University of Victoria. 22.
(28) UVIC HEP Research Computing. Cloud Scheduler Cloud scheduler looks at the job queue Makes a request to boot a user-VM on a cloud The cloud retrieves the user-VM from the repository The user-VM attaches itself to the Condor pool and Condor sends the user-job to the user-VM. The user-VM stays active if there are more jobs that require it. Randall Sobie. IPP/University of Victoria. 23.
(29) UVIC HEP Research Computing.
(30)
(31) . . . !. . . !. Randall Sobie. IPP/University of Victoria. 24.
(32) UVIC HEP Research Computing. Astronomy applications CANFAR Project Canadian Advanced Network for Astronomical Research UIVC, UBC, NRC-HIA CANARIE-funded project Distributed cloud used to process survey data In production for 8 months using different IaaS cloud resources Compute Canada cloud site at UVIC Enabling system for user analysis as well as production jobs See Tuesday presentation Randall Sobie. IPP/University of Victoria. 25.
(33) UVIC HEP Research Computing. Current status • Run up to 500 simultaneous jobs over 8 clouds • > 100,000 successful jobs • Should scale for low-IO applications. • Testing user-analysis (chaotic, high IO) • Early test :10TB analysis in 2 days. • Caching of user-VMs (only transfer 1 16G VM) • Testing squid-cache. • Network challenges • Needed to work with CANARIE to resolve some issues • CANARIE will soon connect with Amazon EC2. Randall Sobie. IPP/University of Victoria. 26.
(34) UVIC HEP Research Computing. Summary • Our distributed cloud focuses on applications in physical sciences with large high-throughput (HTC) workloads and a knowledgeable user community • Fault-tolerant system using multiple-IaaS (commercial or science) cloud resources • Based on open-source components with two new in-house elements • Cloud scheduler and Repoman (VM repository). • Production use by astronomers and BaBar • >100,000 VMs booted. Support provided by CANARIE, NSERC, NRC, Amazon, Google, FutureGrid (NSF) Randall Sobie. IPP/University of Victoria. 27.
(35) UVIC HEP Research Computing. (:+ . $( ( 0 . . ;< .
(36)
(37) .
(38) (0!!) .!( )! +((0!!) .!( )!(. $$( (0 . '2.'75= 8(0!! . '.) 4 ;<.2(0!! . (6.). ;<#."(0!! .. .! *'(0!!) .. .!(' 5(0!!'... ;<
(39) ./.,(0!! 1 ..)! '.((! 9*) 3 .#. &%-/(0!! .3 .)
(40) .".27(( 8(0!! ).). Randall Sobie. IPP/University of Victoria. 28.
(41)
Related documents
Also both negative and positive perfectionism could significantly predict changes in self handicapping (p<0.001). Conclusion: In the presence of positive
In the interview, students were asked about their educational goals; their experiences with searching for, evaluating, and using information; their experiences with IL instruction;
• authentication uses AD when the user logs on using a domain rather than local account. • user’s credential information is sent securely across the network to be verified
This is the first time were direct 24-hour energy expendi- ture measurements in healthy infants with a standardized methodology [6], was used as a reference to test the accu- racy
To address their evaluation behaviors, we asked (1) what strategies students used to read and evaluate scientific literature, (2) how easy or difficult students found
The results of these experiments indicate that methanol may be able to remove PAH’s from field soils effectively; this work furthers knowledge by (1) observing differences in
A short distance in front of the chapel is an old well whose mouth is formed by a large block of white marble (P1. 31,b), and just beyond this are the ruins of
In this case, a proton (from an acid) attaches itself to one of the unshared pairs of electrons on the oxygen.. The carbonyl group now has a +1 charge and is very inviting to even
