Slide 4 of 19
Notes:
individuals file are laid out sequentially - but not together
metadata - attribute/data/directory entry are separate
5 I/O’s to create a file:
2 access to file attributes
1 access to directory data
1 access to directory attribute
< 5% of disk BW utilized in FFS
synchronous writes are bad:
performance becomes disk-bound (waiting for disk to finish)
How about “buffered writes on disk”?
Lots of small writes - overhead dominates (SCSI/bus/filesystem/disk/etc)
