LFS

Context

• Disk
  • Seek time performance is still relatively terrible on hard disks
  • Transfer rate is getting a lot better
  • Capacity increased dramatically (more tracks/disk and more bits/track)
• Memory
  • Capacity increased dramatically → larger file buffer cache → fewer read going to disk, but the same amount of write going to disk → we start doing write operation asynchronously
• Issues with FFS
  • FFS reduced the seek time, but there are still too many seek operations
  • For Inodes and metadata blocks, FFS still needs synchronous writes (cannot exploit larger file buffer)

Goal

Implementation

• Write operations are treated as logs and appended to file buffer in memory.
• When it gets to a certain size, write to disk

Takeaway

My Question