devine@cookie.dec.com (Bob Devine) (01/06/88)
Here's a topic I haven't much of lately -- new arch for Unix file
systems. Here is the timeline that I remember off-hand:
early 1970's : Ken Thompson's original design
early 1980's : UC-Berkeley's Fast File System
These dates don't include AT&T's tweaks to "harden" against
damage or Sun's NFS with vnodes and AT&T's RFS.
There are two issues : speed and efficiency of disk use.
AT&T's sys-V version is still true to Thompson's indirect,
doubly-indirect, and triply-indirect block pointers, though the
block size has changed. Thompson's design gets slower for large
files (and the average file size is growing year by year) but is
hard to beat for disk-use efficiency (one inode + data blocks).
Berkeley's design is better for large files because of much
larger data block size but is less efficient because of internal
fragmentation. UCB's FFS works great for one user but doesn't
maintain speed for N-user, random-banging use.
What new work is being done? Since CMU is getting some DARPA bucks
now, is anything planned for MACH?
Bob Devine