Hi gurus,
here are the aswers to my question
"is it usefull to defragment an advfs domain residing on a RAID
0 or RAID 5 hardware?"
!!!!! YES YES YES YES YES YES !!!!
It is really usefull both for meta-data and for actual-data.
Many thanks to :
-------------------------------------------------------------------------
thomas.erskine-dated-9e70a308e889b1bc_at_silverlock.dgim.crc.ca
Very usefull. I've had them go unrecoverably bad without.
-------------------------------------------------------------------------
alan_at_nabeth.cxo.dec.com (Alan Rollow - Dr. File System's Home for Wayward Inodes.)
Yes, it is meaningful. First, the defragmentation process is able
to collect file system metadata structures is a smaller number of
larger extents, whice helps avoid some of the problems it has with
large number of small files.
In general making other large files contiguous gets the distributed
space close together within the same seek range of the disks. To
take an extreme example. Suppose a file has one page allocated
from the beginning of the first member of the array, the 2nd page
allocated from the end of the same disk, the next page allocated
from the extreme end of another disk and then the last page allocated
from the beginning of the first disk. Reading this file causes
extremely long seeks for the one disk and only a small amount of
I/O from the 2nd.
With defragmentation, the file will end up being contiguous and
will probably end up on a single disk, but can read in a single
I/O. Large files that are badly fragment will have data spread
all over the RAID. When defragmented, they may spread out evenly
across disks.
In a file system that depends on contiguous allocation, having
contiguous data is a good thing.
-----------------------------------------------------------------------------
Dr. Tom Blinn, 603-884-0646" <tpb_at_zk3.dec.com>
Even though AdvFS may not be worried about file placement (and I'm not
saying whether it is or isn't), AdvFS defragmentation is a "good thing"
because it deals with consolidating files into single extents (which can
improve sequential access), and consolidates free space on the domain in
large extents (which reduces file system metadata requirements). Unless
you really need the CPU cycles for other things (in which case be sure
you "nice" the defrag process), it's a good idea to do what you're doing,
so much so that in V4.0D it's a capability included in the base OS that
is enabled by default.
Thanks to dr. BLINN for the further explanation he gave to me.
Now I really understand what I missed :
..........
..........
The actual physical location of a data page in an AdvFS file domain may or
may not be contiguous to any other data page associated with the file in
which the page is located. If you want to increase the probability that all
the pages of a given file will be in consecutive logical blocks on the disks
in the domain (which may only be logical disks if you're using RAID), then
you have to write the entire file sequentially to the domain while no other
activity is occurring in the domain. In this case, it's as likely as it
ever gets that the data will be both physically and logically contiguous on
the disk (that is, in consecutive logical blocks).
........
........
--------------------------------------------------------------------------------
Richard L Jackson Jr <rjackson_at_gmu.edu>
The Advfs defrag utility defrags the Advfs filesystem metadata. Advfs
metadata is a higher level than the RAID device.
---------------------------------------------------------------------------------
Thanks to all the alpha-managers who keep this mailing list alive and
very usefull !!!
Emanuele Lombardi
mail: AMB-GEM-CLIM ENEA Casaccia
I-00060 S.M. di Galeria (RM)
ITALY
mailto:lele_at_mantegna.casaccia.enea.it
tel +39 6 30483366
fax +39 6 30483591
Received on Tue Mar 24 1998 - 11:09:31 NZST