VIEWS: 2 PAGES: 20 POSTED ON: 2/25/2012
FusionGrid enables more efficient use of computing resources • Grid computing is used to securely share computing resources • Complex codes are centralized – Easier for service providers to share their codes with a broader audience – Those best suited to support a complex code—the authors—can do so, and on their own computers • Ease-of-use is emphasized: data, codes, and other resources abstracted into the idea of a grid “service” • Security through certificates and authorization database – A single account makes it easier for users to use grid resources that may be on multiple computers and multiple sites Optimize the most expensive resource: peoples’ time APS/DPP 2004 v 1.1 (20041108) Overview 1/3 The National Fusion Collaboratory is working towards the Collaborative Control Room concept • Next-generation tokamaks will NOT be in the United States – ITER in France or Japan • U.S. leadership in fusion research will therefore require effective use of remote collaboration technology • The National Fusion Collaboratory is working on a collaborative control room – Fully engaged remote scientists with audio, video, shared displays – Secure computational resources that can be scheduled as required – Share individual results with the group via shared displays – Rapid comparison of simulation with experiment APS/DPP 2004 v 1.1 (20041108) Overview 2/3 FusionGrid Computing and Collaborative solutions are actively used in the Magnetic Fusion Energy community • The TRANSP transport analysis code is deployed as a FusionGrid service – Now the official TRANSP installation, centralized at PPPL – Over 4,200 TRANSP runs – High-quality support by authors of the code • Collaborative Control Room concept used & evaluated – San Diego scientist remotely led experiment in U.K. – Subsequently used between ASDEX-U, JAERI • GATO stability code service added – First test uses of this service at General Atomics APS/DPP 2004 v 1.1 (20041108) Overview 3/3 Remote Computing made easier though Globus Toolkit • Codes and other computational resources are abstracted into grid “services”, which are accessed securely • Services provided by multiple hosts at multiple sites, but abstraction hides this from users to make things simpler • Users “sign on” to the grid, not each individual host on the grid – A lot of work to implement this, but makes things much easier for users • Single sign-on implemented through X.509 certificates • Secure MDSplus—built with components from the Globus Toolkit—combines the familiar MDSplus data system with Globus middleware for secure data access APS/DPP 2004 v 1.1 (20041108) Authentication made simpler through X.509 certificates • Each grid user has a single identity for the entire grid – Implemented as an X.509 certificate – Usually kept in two files: a certificate and a private key • A centralized Certificate Authority (CA) serves as the authority for creating or revoking certificates • For ease-of-use, users now have the option to store their certificate on a MyProxy server – Retrieve using a username and password – No need to copy certificate files around – For security reasons, needed to create a CA for FusionGrid to implement this APS/DPP 2004 v 1.1 (20041108) ROAM developed for centralized authorization and account mapping • Centralized authorization was needed so users can easily request authorization to use resources – Also makes it easier for administrators to manage grid-wide security • Resource Oriented Authorization Manager (ROAM) developed to manage user permission on FusionGrid • Permissions can be set for each resource on FusionGrid – Each site (e.g. General Atomics, MIT) is modeled as a resource, too • The mapping of grid identity to local accounts is done through this system – Before ROAM, this was problematic • Mapping can be as specific as needed, down to the individual resource • Both users and programs access database through secure web server – Users can load their certificate into their web browser, or use a username and password APS/DPP 2004 v 1.1 (20041108) Resource Oriented Authorization Manager (ROAM) APS/DPP 2004 v 1.1 (20041108) Grid resource monitoring improved through FusionGrid Monitor (FGM) • Abstracting computing resources makes it more difficult for users to monitor the progress of their computations • The FusionGrid Monitor (FGM) was developed for efficient resource monitoring • Resources post monitoring information to the FGM server • Monitoring data is pushed to web browser clients using server push technology – Much more efficient than polling (e.g. frequent refreshing of browser) – Client pull available for browsers that do not support server push • Logfiles available through web interface • Data persists in a relational database – Useful for analyzing performance APS/DPP 2004 v 1.1 (20041108) Collaborative Visualization using ElVis • ElVis Java applet deployed to visualize TRANSP data through FusionGrid Monitor (FGM) • ElVis can be used collaboratively so multiple researchers can view and annotate the same data remotely – Share data and interactive visualization. – Verify certificate credentials of collaborators. – Maintain firewall security. • Can retrieve data directly from MDSplus APS/DPP 2004 v 1.1 (20041108) Access Grid provides immersive remote collaboration • Access Grid (AG) connects remote scientists through audio, video, and shared applications • Works over the Internet – No special lines required • Scalable: entire rooms for large groups, small personal nodes for individuals or small groups – All are interoperable, so lone researchers are free to collaborate with large institutions using the same technology • Open source, extensible APS/DPP 2004 v 1.1 (20041108) ShareD: Software for Application Sharing • Display information sharing – Fusion scientists can move computational results from their desktops to the shared display. – Visualizations from all scientists can be compared side-by-side. • Concurrent Control – Scientists can simultaneously edit and interact with applications. – Researchers can annotate visualizations simultaneously. – Access and edit control are policy driven and configurable. Collaborative APS/DPP 2004 Users Shared Display v 1.1 (20041108) SCIRun used for advanced visualization of fusion data • SCIRun is a problem solving environment – Visual programming interface for power users – Advanced visualization • Can be run through FusionViewer application • Reads MDSplus and HDF5 data • Open Source APS/DPP 2004 v 1.1 (20041108) Web Portals provide an easier interface for users • Instead of asking users to learn new applications, web portals provide a unified interface to grid resources • A lot of work to implement, but makes it much easier for users to use grid • Currently testing a web portal for TRANSP APS/DPP 2004 v 1.1 (20041108) New interface between MDSplus and SRB • Storage Resource Broker (SRB) now works with MDSplus • Extends the number of interfaces to MDSplus – Access the data via web services • SRB brings many capabilities – Can store data into HPSS and other archives – Adds tools to easily replicate data into multiple resources – Higher performance data transfer (via parallel I/O) – User-defined metadata capability to locate datasets – The separation of the logical from physical name space allowing for multiple representations of the collection hierarchy – Access to preservation environment for managing technology evolution APS/DPP 2004 v 1.1 (20041108) Storage Resource Broker and MDSplus MDSplus Client SRB Interface MDSplus Server SRB Client SRB Switch SRB SRB SRB Local Disk SRB Servers (connected to HPSS, Web APS/DPP 2004 v 1.1 (20041108) Portals, etc.) Advance Reservation feature being developed to allow scientists to reserve computing power • Computing power can be reserved ahead of time • Most useful for experiments where rapid results are critical • When needed, computing power is “claimed” by the scientist • Advance reservation uses “agreements” to negotiate CPU reservations – Other reservations are possible, not just CPU • Possible to “renegotiate” agreements if needed • When claiming computing power, other running processes are lowered in priority while scientists’ processes are raised in priority, making full use of CPU APS/DPP 2004 v 1.1 (20041108) Advance Reservation 1. Negotiate 2. Create Agreement 3. Renegotiate/Manage (optional) Agreement negotiation 4. Claim Agreement Factory agreement creation 1 Agreement Management Agreement Entries: renegotiation 2 Agreement Terms 1 Client … 3 agreement management Agreement Terms N 4 Service APS/DPP 2004 v 1.1 (20041108) GATO FusionGrid service deployed • GATO stability code available as a FusionGrid service • Runs on dual-CPU Linux computer at General Atomics • PreGATO client used to launch GATO – Simple run management – Authorization query capability – Can retrieve GATO output files (useful for debugging) • GATO outputs stored in MDSplus • GATO run management through Code Run Database • FGM for monitoring • ROAM for authorization APS/DPP 2004 v 1.1 (20041108) GATO FusionGrid Service ROAM FGM Client job submission→ Server GATO PreGATO MDSplus GATO tree Code Run DB APS/DPP 2004 v 1.1 (20041108) Continued success of TRANSP service • The FusionGrid TRANSP service is the official way to run the TRANSP code. • Almost 3,000 runs completed • Multiple tokamaks • The first FusionGrid service and a continued success 3000 2500 through 2000 10/31/2002* 1500 11/1/2002 -- 10/31/2003 1000 11/1/2003 -- 500 10/31/2004 0 Total Runs APS/DPP 2004 v 1.1 (20041108)
"Slide 1 - Fusion Grid"