GridPP20 Collaboration Meeting_Dublin ============================= 11.03.08 - 'Users' Discussion Session (Chair: Pete Clarke) ------------------------------------------------------------------------------------ PC began by noting how important it was to get the 'user' impression of GridPP, and asked how many users were in the room. A low number was present. PC asked how many of those present were likely to be users by this time next year - around six additional people responded. PC commented that 'users' usually meant experiment reps, and asked whether we should have more users here next year. RJ was asked to give comments on how the UB views the Grid. RJ noted that the UB discusses resources and will go on doing so - the UB 'view' was therefore the 'resource' view. RJ commented that James Catmore has had experience of both: doing user analysis plus the production role for physics groups - he would say that the Grid was good now for production level work; he would also say that it was a generally good experience of getting work through the Grid, but that it was fragile - there were problems accessing data at sites. RJ reported that Dave Charlton (ATLAS FDR-1 Co-ordinator) had asked how people had managed - they go and copy the data, and this is not good. There is therefore an education effort needed to persuade users to use the Grid, but we would need to block Tier-1 access to the user and also make the user experience better - it is still too difficult for users to use the Grid. The question was asked: why can't users get access to data at RAL? Greig Cowan noted that data could be anywhere: CNAF, CERN etc. RJ advised that ATLAS users shouldn't use RAL, but he noted that it also was not obvious to a user if the data was on tape or disk. Raja Nandakumar commented that the user doesn't need to know that. DC advised that CMS was different, and user analysis was largely Grid-based. A delegate disagreed with this, saying that it was not the case during CCRC'08. PC noted that if people use 'back-door' methods then was it our job to do something about it, or was this the experiments' responsibility? GS noted that GridPP ensure that systems work well enough so that analysis jobs work well for users - experiments themselves need to be the ones to 'police' users. RJ advised that tools to police users don't exist. PC asked what the fundamental issue was, and suggested that it related to catalogue, store, etc. GS reported that when users get error messages they assume the Grid has crashed. RJ commented that physicists need to know something about the system itself in order to use it effectively. Brian Coghlan noted that if a user does have a problem they simply click 'retry'. A delegate commented that if a job fails at one site then it can be resubmitted elsewhere. RN suggested that users were patient as long as the documentation was up-to-date - there was usually not a problem in his experience - users can wait for 2-3 weeks for their jobs to come back. PC commented that people put up with a great deal now because data-taking has not started - but once discoveries are possible, patience levels will go down, and will GridPP be able to work with sharp-end users who demand immediate results. DC noted that we already work with users to enable them to be as effective as possible, but they generally don't attend Collaboration Board meetings. RJ advised that a lot of this support goes through the experiments and their attempts to set-up more local support. DC noted that local complaints were the first line of support. PC suggested that we should try and encourage users to come to the next Collaboration Board meeting? Alessandra Forti commented that there should be sessions at the experiments' national meetings. DC advised that CMS do have meetings. Jens Jensen commented that NGS have asked the same question, and managed to get users together and share their experience in a 'users' forum', but very few generally turn up; they also have 'user of the month' published on their web pages in an effort to build user communities. RN noted that experiments have different ways of approaching the Grid, therefore common education of users was difficult. SP noted that EGEE had addressed the same issue - but the EGEE user forums remain largely unattended by users. NG advised that users don't come to user fora - they typically talk to their peer group. What would be most effective, he suggested, would be to take a road show to users' meetings - there was a feeling that GridPP was rather detached from users in general. PC summarised by saying that the PMB would discuss this issue and take it forward. All participants were thanked for their contributions.