CRAM, DataCell, and 3850 4017

re:

we had instrumented several systems in the san jose area, so that we could get trace of every record accessed. we used it for some i-o (electronic memory) caching models ... comparing effects-sizes of disk arm level, controller level, channel-bus level , and system level record-track caches.

one of the interesting things that came up of that study was there was finding quite of bit periodic activity that used collections of files ... and in fact, some of the collections had very strong affinity (i.e. they were only used when others in the collections were also used). conjecture was that it was related to periodic (frequently quite deterministic) end of specific business applications. some amount of this was quite large sequential access (much larger than most traditional electronic memory "caches" ... i.e. the standard electronic memory caches provide little or no benefit for such activity)

you see some of the backup-archive systems (like TSM) attempting to optimize for such behavior with things that some refer to as "containers" ... i.e. groups of files that are managed as a collecton ... where contents of containers may also be compacted-compressed as a unit.

for other drift ... the current generation of tapes can have such large capacity ... there are deployment considerations given to moving stuff first to disk-based "virtual tape reels" ... and then "stacking" large number of virtual tape reels from disk to a single large capacity physical tape (some of the traditional backup-archive to tape software tended to use a minimum of a single tape per operation, which led to frequently leaving new generation of physical tapes almost entirely empty).

misc. past backup-archive relates posts

misc. past references to file-disk activity traces-modeling: