CERN Accelerating science

During the last twelve months we have been executing a plan to modernise and homogenise the AFS service both at the hardware and software level. By the end of May we had replaced some dozen old file-servers, a mixture of IBM and Sun having disk spaces ranging from 50 to 150GB and not all on secure RAID, by 5 Sun E220 servers each with 300 GB of RAID disk. In June we replaced the three old volume location data base servers by new Sun Netra servers (which no longer have to function also as file servers) and in July the addition of a sixth Sun E220 allowed us to shut down 14 old PCs running a non-standard version of AFS. We also added a large Sun E450 machine to function as a central backup server using an STK DLT 7000 robot and replacing a number of directly attached DLT stackers. All the new machines have fast networking.

All the remaining or new servers are capable of running the same latest level of the AFS server software namely 3.6 patch level 1. This level is multi threaded, so more efficient, and supports backup and restore of volumes up to 8GB in size (the previous limit was 2GB). We successfully migrated to this level on 31 July. When this has proved stable AFS client machines can be brought to the same level.

The final changes planned for this year will be to replace the last two IBM file servers, which use our deprecated FDDI networking, by two more of our standard new configuration. This should happen during August or early September and should be transparent to end users.

We appreciate the patience of our end users during this evolution which has already led to improved reliability and performance. Changes next year should be much more transparent. There are two older Suns that should be replaced then new capacity to be added according to the COCOTIME evaluations currently underway.

The changes have meant that all end user home directory and backed up project directory files are on secure RAID disks. Two disks out of six can fail without losing access (there is a hot spare mechanism). The non-backed up project directory files are also on modern disks but a single disk failure will lose data. It is very important that groups only use the non-backed up space for genuine scratch data.

You can recognise which type of disk a directory is on by the command

fs lq your-directory-name

user.     this is a backed up user home directory
u.        this is a backed up user workspace directory
p.        this is a backed up project directory
q.        this is a non-backed up project directory - SCRATCH ONLY