When the processed data reaches the size of an exabyte tape, it should be backed up. Moreover all files should be backed up twice in case a tape goes bad. This can be accomplished in two ways, either the same backup is made on two tapes, so there are two identical copies of all data. It is also possible to split the data in two data banks, each with the size of half exabyte tape. With the first method you could delete all the data after the two backups have been made and thus free about 3.6 Mega blocks on the disk. If the second method is used, only one backup is made and one of the banks is deleted and the other is saved for the next backup. This means that you only free 1.8 Mega blocks on the disk with this procedure.
The first method has the advantage that only one backup reference file is needed for the two tapes(since they should be identical). It also implies that backup only needs to be done with half the rate as compared with the second method(with the drawback that the backup takes twice the amount of time as two tapes need to be backed up). The second method has the advantage of producing a more stable disk space level on the disk, that is changes of 1.8 instead of 3.6 Mega blocks.
As the databanks are resident on the same disk as the incoming data, a stable disk space level as possible is desirable. When the DELANA version is changed a reprocessing of the old data starts, in parallel with the new incoming data. There is a quick buildup of data files and loss of disk space. It is therefore also an advantage that less data is needed to fill up a tape and backups can be done more quickly and often. If errors are discovered it is also desirable to have the latest DELANA files on disk, in case reprocessing of a fill is needed. For all of these reasons THOR uses the second method mentioned, even if this might be a bit of extra work for the shifter(not much though). There are mainly three types of data that should be backed up:
If we organize this into two banks, we get the structure seen in figure 1. When the size of bank1 is bigger than 1.75 Mega blocks a backup is made. The backup is done on both bank1 and bank2 to fill up the exabyte tape. After this backup, bank2 is deleted and all the files in bank1 is moved to bank2. The newly processed files will then start filling up bank1 again and so on. As more disks have became available the FX-files are now backed up separately in their own data banks.
The exabyte tape cannot be loaded automatically and the shifter is thus informed by mail when bank1 is full. He should then insert a tape and confirm this after which a backup and cleaning procedure automatically is started. When the data has been backed up and cleaned away the shifter once again receives a mail that he should remove the backup tape and write the proper name on it.
