Presentation on theme: "Andrew McNab - Manchester HEP - 24 May 2001 WorkGroup H: Software Support Both middleware and application support Installation tools and expertise Communication."— Presentation transcript:
Andrew McNab - Manchester HEP - 24 May 2001 WorkGroup H: Software Support Both middleware and application support Installation tools and expertise Communication (web, mail lists) Documentation and software repositories Contacts with Globus Middleware API definitions Management of Grid partitions Getting it all to actually work...
Andrew McNab - Manchester HEP - 24 May 2001 Can we fix it? Yes we can! (Subject to available resources …)
Andrew McNab - Manchester HEP - 24 May 2001 Installation Tools Three layers of software: –a. Operating system - automated tools exist –b. Grid middleware - eg our RPM distribution –c. Applications - currently, the experiments all use different distribution methods a / b / c dont need to use same mechanisms, but they must work together. Need to be able to specify dependencies in the lower layers.
Andrew McNab - Manchester HEP - 24 May 2001 Interfaces with other groups Overlap with WorkGroup D (Fabric and Data) Need to agree package formats that will plug into fabric management systems. EU DataGrid WP4 (Fabric) is currently defining installation formats for EDG Testbed 1 - clearly this will be a major influence on LHC experiments policies. What about Babar / CDF / D0? Clearly, we need a flexible standard.
Andrew McNab - Manchester HEP - 24 May 2001 Installation proposal Carry on using RPMs –Support for source and binary packaging –Dependencies –Natural for Redhat Linux etc. –Portable to Solaris etc RPM is where WP4 seems to be going. Where necessary, use dumb tools to translate tar files to RPM - this mostly applies to HEP applications.
Andrew McNab - Manchester HEP - 24 May 2001 Support for installation etc WG-H provides support to local administrators - ultimately helpdesk? This means that we develop of pool of expertise within WG-H about common problems which everyone can draw on. Important to communicate bugs, design faults etc back to Middleware and Application authors.
Andrew McNab - Manchester HEP - 24 May 2001 Communication Tools Website - www.hep.grid.ac.uk –Allows write access using Grid authentification. Mailing lists - ukhepgrid and ukhep-globus In other Open Source projects, WWW and lists allow users to aid in development. We should implement searchable list archives on the web, and make it easy for subgroups to start their own lists.
Andrew McNab - Manchester HEP - 24 May 2001 Documentation Need formal documentation if the Grid is to become pervasive (ie not just something for experts.) This is also something to offer for the Dissemination effort. Call these Guides? XYZ User Guide, XYZ Installation Guide etc. But also need informal HOWTOs for experts, which can keep up with rapidly changing code.
Andrew McNab - Manchester HEP - 24 May 2001 Software repository A central place where code (and documentation) can be stored and shared. Access via CVS and Web. Nightly builds? Problem reporting to module contact person? What about copyright issues? Follow same practice as Globus: if you contribute patches you are granting a COGS-compatible license?
Andrew McNab - Manchester HEP - 24 May 2001 Contacts with Globus Globus have opened up their internal developer lists and CVS. This means we can already participate in their development process to help with things we need. Weve also made direct links, partly via our participation in EU DataGrid. Globus appears very open to external collaborators, and the best way to get specific work done may be to send people to ANL or ISI to contribute effort?
Andrew McNab - Manchester HEP - 24 May 2001 Upper Middleware API WG-H also has to draw together the various APIs from other WorkGroups to specify the environment available to Applications. By agreeing common APIs for access to things like meta data, other WorkGroups can produce general components (eg super schedulers) with arent tied to a specific experiments database structure.
Andrew McNab - Manchester HEP - 24 May 2001 Management of partitions Especially near the beginning, were going to have incompatible requirements on what is installed. Short term solution is to divide the prototype Grid into appropriately configured partitions. Should bear this in mind when deciding on fabric management and installation mechanisms. Want to be able to reinstall machines in a partition in an automated, reliable way. In principle, this can be done in a few minutes.
Andrew McNab - Manchester HEP - 24 May 2001 Summary Most of the component activities of WorkGroup H have already been started: –Installation tools (RPMs) –Web site –Mailing lists –Contacts with Globus But documentation, co-ordination of APIs and active support are still just starting.