evals for 6/1/00

Aaras P VASA (avasa@cs.ucsd.edu)
Thu, 1 Jun 2000 10:22:14 -0700 (PDT)

a fast file sysytem for Unix

this paper reimplements unix file system to one that provides two block
sizes to allow fast access to large files and not wasting space for small
files

traditional systems using 512-byte blocks fail to provide higer bandwidth
for
large files.they incur many seeks and have limited read-aheads
in the FFS block size is changes to 4096 bytes
the file system organisation divides disk partition into cylinder groups
which contan bookkeeping information. to optimize storage
file system block is divided into fragments

this new system outperforms the old system for sure.
this was a great idea and along with this the other enhancements
like file locking, symbolic linls were added .
the overhead of allocating blocks is greater in this system
but is not a big worry
i think this was a great design at teh time

the design and implementation of a log-structured file system

this paper presents the log-structured file system which writes
modifications to disks in a log-like strucure.

appliocations are becoming disk-bound and disk traffic is dominated
by writes.also applications tend to be dominated by small files
this is why log-structured file systems were implemented
they buffer a sequence of file system changes in cache and write
them sequentially.
LFS writes inodes to logs and divides disks into segment

i think LFS is a better option to UNix FFS
crash recovery is easier in LFS with checkpoints.
the main assumption which should hold in enough cache for files
and mostly small writes . if not the FFS might be better as
LFS is not conclusive about it.
the cost-benefit policy seem to solve the main overhead of segment
cleaning.