CASTOR at CERN provides the service for storing large volumes of physics data (RAW data, simulation files) and user files. Details on the Service Level agreement are provided on this site. Physics data are stored on a disk pool and then copied to tape, the disk copies are deleted asynchronously (typically to avoid to fill up the disk completely). User access to these data (especially tape recall to create a disk copy for data residing only on tape) are regulated by the experiment/user community policies (please refer to the SLA section on the documentation page). Efficient writing is achieved by with large files (1 GB and above). Good read performances are achieved for disk-resident data (or pre-staged on disk for you by your experiment). Files should have a tape copy within 24 hours (8 hours for physics data). The tape recall delays have large variations and they are often discouraged by the experiment community. As a guideline, 1 hour delay is a typical value: during this time your programme will simply wait and data will be processed only a successful recall will be completed. In general users files (like ntuples for analysis) go to disk but not to tape (these are non custodial data, i.e. data could not survive some hardware failures). Space management is entirely with users (and corresponding user community): in case of "disk full" no further writing is possible until obsolete data are removed by them. Users should use this to analyse large files (typically skims of experiments files). For moderate volumes you are encouraged to use the AFS service.