330 likes | 430 Views
The new EGRID infrastructure. An update on the status of the EGRID project. The new EGRID infrastructure. The EGRID project: To implement Italian national grid facility for processing Economic and Financial data.
E N D
The new EGRID infrastructure An update on the status of the EGRID project
The new EGRID infrastructure The EGRID project: • To implement Italian national grid facility for processing Economic and Financial data. • Underlying fabric on top of which partner projects develop Economic and Financial applications.
The new EGRID infrastructure Summary: • Original user requirements • The first EGRID release • Operating problems • Redesigning EGRID • A web portal to access EGRID
Original user requirements • HW infrastructure to store+manage 2TB Stock Exchange Data: NYSE, LSE, Borsa di Milano, etc. • Privacy: legally binding disclosure policies • Users do not have the same read rights: a research group has contract with NYSE for a specific company; another group has contract with LSE for all companies; etc. • Two classes of users: those that upload stock exchange raw data and that may remove it; and those that work on the data. • Facility organised for raw data pre-processing and end-user applications.
The first EGRID release Meeting the HW infrastructure requirement: • Bulk computing power access and bulk storage rented from INFN Padova, part of Physics grid! • Employed same EDG middleware INFN uses. • Two tiered topology dictated by network connectivity: • Partner projects have limited connectivity: installed peripheral sites supply local services. • Cache area for large data transfers. • Job execution points for non CPU intensive data processing. • INFN Padova has good connectivity: supplies services to whole community.
The first EGRID release Meeting data privacy: EDG’s data access mechanism implied critical and fragile fine-tuning. • Classic SE: local files exposed through GridFTP. • GridFTP allows file manipulation compatible with underlying Unix filesystem permissions. • The underlying filesystem must be carefully managed: • Users mapped to specific local accounts: not pool accounts. • Users partitioned into especially created groups: reflects data access patterns. • Carefully crafted directory tree guides data access. • Users have same UID across all SEs. • Replication/Synchronisation of directory structure across all SEs. • Users supplied with tools to manage permissions coherently across all SEs.
The first EGRID infrastructure Meeting pre-processing requirement: supported with tailor made wrapper component. • Developers can more easily grid enable pre-processing operations. • Users to more easily run grid pre-processing on given datasets. • Common Unix commands such as cat,cut and grep, were adapted to operate on grid stored files.
The first EGRID infrastructure Meeting user needs: • User applications are specific to research interests: programmes and function libraries developed to aid porting of applications. • To facilitate installation of grid client SW, LiveCD technology was employed.
Operating problems HW infrastructure: • Only one large computing site: insufficient to demonstrate grid potential for distributed resource allocation. • Two tiered topology problematic: maintenance task on designated local user + EGRID could not dedicate enough manpower to job.
Operating problems Privacy: • EDG and successor middleware LCG still lacked data access mechanism strong enough for EGRID. • Implemented solution is complex and does not scale: real account for each user in each SE, permissions on filesystem make tree replication tricky, etc… The middleware did not allow a solution in line with a pervasive grid view.
Operating problems User needs: • Only small part of community used tailor made command line tools. • UI distributed on LiveCD spared users workstation reinstallation, but: • users complained of awkward usage • interference with usual way of working
Redesigning EGRID • Driving factors: • Leaner and more general infrastructure • Robust privacy • Thoroughly re-examined grid usability
Redesigning EGRID HW infrastructure: • Added second large computing centre: INFN Catania. • Dropped two tiered topology.
Redesigning EGRID Privacy • Classic SE replaced with specific implementation of Storage Resource Manager (SRM) protocol currently being completed. • Implementation is result of StoRM collaboration with INFN-CNAF. • Not a proprietary solution – SRM becoming standard for grid disk access: security solution compatible with mainstream grid trends.
Redesigning EGRID • How StoRM solves privacy: • All file requests are brokered with SRM protocol. • When StoRM receives an SRM request for a file: • StoRM asks policy source for access rights to: given file for given grid credentials. • Check is made at the grid credential level: not local user as before! • Physical enforcement through JustInTime ACL setup: • All files have no ACLs setup: no user can access files. • Local Unix account corresponding to grid credentials is determined. • ACL granting requested access set up for local user. • ACL removed when file no longer needed. • StoRM leverages grid’s LogicalFileCatalogue (LFC) as policy source: compatible with mainstream grid trends
Redesigning EGRID • Completing data privacy: • ELFI tool developed to allow classic POSIX I/O software interface access to grid files. • ELFI is FUSE filesystem implementation: grid resources are seen through local mount points. • ELFI speaks SRM protocol: there is lack of SRM clients.
Redesigning EGRID • ELFI allows more: • All existing file management tools work automatically with grid files: • Text tools: cat, grep, etc. • Graphical tools: Konqueror, etc. • Helps RAD/Prototyping: developers not got to learn new APIs when porting applications. • Sites supporting ELFI on WNs: applications spared need to explicitly run grid file transfer commands.
Redesigning EGRID Grid usability: • Web portal key solution: portals long proved to be effective ways to allow user interaction with organisation’s information system. • Old command line tools will remain: • For backwards compatibility. • For few users that eagerly adopted them. • New development will concentrate on web portal.
A web portal to access EGRID • Main entrance to new EGRID infrastructure. • All tools in one place + Graphical UI: • Closer to users’ way of working. • Lowers resistance to new technology. • No need to install grid SW on users’ workstation: • Interaction through portal as displayed in web browser. • P-grade chosen as portal technology: • Sufficiently sophisticated as starting point to meet EGRID requirements. • Does not fully meet EGRID requirements: extra development needed.
A web portal to access EGRID P-grade’s GUI simplifies many routine task and masks complexity: • No need to manually handle job identification strings. • Display keeps track of launched jobs, status, allows output retrieval, job cancelling, etc. • Easily choose Broker for automatic job submission or specific CEs. • Enough flexibility to allow direct jdl attribute specification. • Graphical browsing of grid resources + file management: no need for distinct tools.
A web portal to access EGRID P-grade portal adds new functionality: • Although MPI jobs can also be run from the CLI, P-grade supplies a special API that allows a graphical report on such jobs to be displayed. • Workflow manager: • Graphically specify several jobs. • Define connections among them showing data flow. • Portal takes care of retrieving job output and feeding it to linked jobs. • Monitoring of workflow done graphically showing data flow.
A web portal to access EGRID Extra development needed: • Improved proxy management • SRM data management • SRM support in Workflow • Support for special workflow jobs: swarm jobs
A web portal to access EGRID Improved proxy management: • P-grade first uploads user’s private key into host where Portal resides – then transfers it to MyProxy Server. • To lower security risks EGRID needs key to be transferred directly from user workstation to MyProxy server. • Java WebStart application developed by EGRID and seamlessly integrated into P-grade credentials portlet.
A web portal to access EGRID SRM data management: • P-grade allows browsing of files: in classic SE + files local to user workstation. • P-grade: does not support SRM + does not support browsing of files in portal hosting machine. • ELFI allows access to StoRM through local mount point. • It is easier to write a portlet that allows browsing of portal local resources rather than one that deals with the new SRM protocol. • EGRID developed a new portlet to allow such browsing.
A web portal to access EGRID SRM support in Workflow: • Workflow definition requires for each job to define input and output files. • For each file must be specified respective location. • P-grade supports: classic SEs + user workstation. • SRM is not supported. • New file location support in P-grade: host containing portal itself… StoRM will be accessed through ELFI local mount point! • On going collaboration with P-grade developers to better define requirement and study feasibility.
A web portal to access EGRID Swarm Workflow jobs: • Swarm jobs: application run repeatedly on different datasets + final job collects results and carries out final aggregate computation. • Currently P-grade workflows allow only manual job parameter specification: automatic mechanism needed. • This feature is already present in P-grade’s release schedule.
A web portal to EGRID Possible drawback: • Java technology is used extensively also on client side: Applets and JavaWebStart used for certain operations… users must have a Java Virtual Machine installed. • Given ubiquitous nature of Java… should not be a big problem.
Acknowledgements • StoRM collaboration with INFN-CNAF of grid.IT project: Dr. Mirco Mazzuccato, Dr. Antonia Ghiselli. • P-grade team headed by Prof. Peter Kacsuk of MTA Sztaki Hungarian Academy Sciences • EGRID project leaders: Dr. Alvise Nobile of ICTP, Dr. Stefano Cozzini of INFM Democritos. • EGRID team: Alessio Terpin, Angelo Leto, Antonio Messina, Ezio Corso, Riccardo di Meo, Riccardo Murri.