Summary: VxFS vs. UFS logging

 Original post:

VxFS vs. UFS (logging)....thoughts, experiences, comments??

I have come to the conclusion that the difference is minimal ( except $$ ).
Both provide the same functionality and basically, performance.
Can the ability to 'defrag' online make that much difference?? I think
not...unless we are talking huge files, large reads/writes...

And then there is the EMC Symm with 16GB cache....again, where is the VxFS
benefit??


The responses are pretty much unanimous, all in favor of VxFS. After
reading all of them, I am convinced that VxFS
is the way to go in our environment. The advantages are RAS, performance
and functionality. Cost is minimal.  A sample
of some of the postings.

VxFS is a journaling file system.  This means it continually
updates the state of the physical disk inodes.  In a crash, all
of the /O data is stored in a "jounal" (hence the name).  Journaled
file systems, are much more (some documents say as much as 900%)
resistant to crash based failures, than regular file systems.
This is the main reason to use VxFS.  The second reason, is you can
customize I/O parameters, like the cached file sysem RAM, buffer,
block size, etc,... in ways you can't modify the UFS parameters.
Finally, Logical volume management.  Ever need to mirror a RAID?
How about chnage the size of a logical volume by adding another physical
disk?
You can't do any of that with normal UFS.

#######################################################################################

If you have a smaller system that is not under much load, I couldn't
make a good argument for paying for VxFS.

If you have a larger system though, the performance tends to improve
quite a bit.

1) No triple indirect pointers.

   Very large files are accessed via extents and offsets within.  Under
   UFS, this can require the read of 2 or 3 data offset blocks to find
   the actual data.  If you're doing this with random data reads (and
   small amounts of data per file), then this can be a significant
   source of possible performance improvement.

2) No QuickI/O.

   Many Database admins enjoy the benefits of specifying individual
   files within a filesystem as available for raw disk access.  In a
   very random access environment, that can be a performance increase.

3) Ability to shrink filesystem without dump/recreate/restore.
4) Ability to create 200GB filesystem in under 30 seconds.
5) No fixed number of inodes.

Any or all of the benefits may be worth nothing to some users, but the
performance benefits are definitely present at high loads.  Whether the
cost of VxFS is worth it is a decision made by individual organizations.

I don't see a EMC (or really any) external cache being much of a big
difference between filesystems.  Most UNIX machines cache aggressively,
so the read cache on an external array doesn't get used much.  The write
cache is necessary, but benefits both VxFS and UFS.

######################################################################################3


Many thanks to the following people. I appreciate the time you took to
respond.

Doug Otto
Ray McCaffity
Mark Neill
Kevin Buterbaugh
Darren Dunham
Paul Frederiksen
Bryce Ryan

Thanks again managers!

Buddy DeMontier
State Street Global Advisors
Infrastructure Technical Services
2 International Place
Boston Ma 02110
617-664-6141