Hello, I'm interested in one of the more boring aspects of high throughput cluster computing which is that of implementing chargeback schemes and organizing queue systems to insure, for example, that a user who might have bought %40 of a cluster (in terms of nodes) is always assured priority access (on 40% of the nodes). Now. I use Sun Grid Engine and have used it to establish priorities and queues though I'm curious how other people manage this problem. Users who actually buy nodes to add on to the cluster are particularly conerned about being able to use "their" nodes whenver they submit a job. Of course I explain and show them to them how the queuing system (Sun Grid Engine) works and while that helps them to understand what is going on underneath the hood its a common problem that I have to deal with perceptions (mostly from power users) that they can indeed get "their share". How are others setting things up in terms of policies ? Are you creating subscriptions based on anticipated usage ? Weighted queues to reflect someone's investment in the cluster ? A straight , one-class queue system ? Prime vs non-prime ? Post job accounting records for chargeback ? I've come to understand that many academic centers are blissfully free from having to do any type of chargeback (lucky people indeed) though some aren't. Any one want to share their views and experiences ? Steve Pittard | http://www.bimcore.emory.edu Emory University | wsp at emory.edu, wsp at bimcore.emory.edu BIMCORE Support | Phone +1 404 727 0038 FAX +1 404 727 5512