Data Storage: REST vs. POSIX for Archives and HSM - Page 3
Ask yourself, when was the last time the industry had 10 announcements about file system in a quarter? My bet, it was the late 1990s and today we get 10 announcements about cloud interfaces and alike sometimes in a week. REST and REST-like interfaces are gaining massive momentum, but from the vendors I have looked at they are almost without question all missing one thing, and that is tiered storage, where an archival policy can be set.
Of course there is Glacier. But when my next relative dies and I want all the data back I do not want to have to pay a fortune or figure out what I can bring back what day as compared to what can be done with POSIX HSM today.
From the 1980s to about the 1990s, POSIX file system world lived without HSMs and users had to figure out how to archive their data. Mostly during this time period mainframes operating systems, which had HSMs, managed those tasks and it was difficult moving things back and forth.
When file systems vendors added HSMs to their POSIX file systems the world changed. Users wanted to integrate into the large file systems, and wanted to move off the mainframes to have computation and data closer to their applications. The trend in the 1990s archive neared the computation, but this changed again as disk drive capacities increased significantly and many did not need archives.
This trend changed again in the late 2000s, when the disk drive capacity trends started to drop and tape, at least enterprise tape capacities, increased tremendously compared to disk. HSM and archives (aka tiered storage) now made significant financially sense to all.
About this time is when REST interfaces became very popular, but the developers took the replication and spinning disk approach, given disk drive density. It took a good 10 years after the POSIX file system under UNIX before someone developed an HSM to tier data using a POSIX file system. My bet is that history will repeat itself for REST and clouds but a bit quicker this time, and that we will see automatic tiered storage.
My hope this time is that standard policies and interfaces will be developed, which was missing from the POSIX interface and prevented, in my opinion, broad adoption.