Loosely speaking, one way to tell the theoretical maximum you can store on a
file system is to do a "df" command.  If you add the -i parameter, that will
tell you the number of inodes you have available.  The maximum number of
files will be equal to the smaller of the 1K blocks and the inodes.  If you
bump up the number of inodes, you will increase the amount of file system
overhead, but also increase the number of files that can be stored.  When
the number of inodes >= the number of 1K blocks, inodes will no longer be
the limiting factor, and the number of 1K blocks available will be the
number of files you can store on that file system.

Note that _none_ of this has anything to do with the size of the underlying
blocksize that the DASD was formatted with.  That is another variable,
independent of this bit of discussion, but relevant in that it determines
the maximum number of 1K blocks you start with, before you start subtracting
for inodes, etc.

The same will be true for 4K blocksizes (just divide by 4 after doing the
mke2fs).

The fact that you ran out of inodes means that if you re-ran the mke2fs with
a different inode value, you would be able to store more files on that file
system.  Proper selection of inode values frequently isn't an issue, but it
can become one, as in this case.

The fact that unused space showed up in your "df" commands means that the
space was there to be used, but couldn't be used because an inode wasn't
available to point to it.  It does not mean that "ECKD overhead" made the
space unavailable to start with.  _That_ overhead gets "subtracted" before
Linux gets to see the DASD.  Formatting DASD with various blocksizes can be
compared by doing a "cat /proc/dasd/devices" and using the "active at
blocksize: 4096, 600840 blocks, 2347 MB" data between the different runs to
compare.


Mark Post

-----Original Message-----
From: Linux on 390 Port [mailto:[EMAIL PROTECTED] On Behalf Of Jim
Sibley
Sent: Monday, March 22, 2004 12:39 PM
To: [EMAIL PROTECTED]
Subject: Myth of the 1K blocksize on eckd (corrected)


To answer both Mark and Tom

1) Both the 1k and 4k block experiment topped out at
54% of the file system full, so, yes, you do run out
of inodes sooner, but you have more "capacity" for
inodes with 4k blocks.

Additionally, chasing the inode chain from disk (not
cache) would be quicker with 4k blocks because there
would be 1/4 the number of i/o's.

2) No, you do not get more blocks with 1k blocksize
because of the heavy eckd overhead.

          1k              4k
blocks  1,626,740       2,366,164

And fewer blocks are used to contain directory
information in the 4k case.

If you increase the number of inodes, you will still
get more small files on a 4k eckd image.

Tom is right in that, if some of your files are more
than 4k, then 4k rapidly becomes the choice, even if
you are not inihibited by eckd formatting. Even if
/etc, which are "small files", most of the files are
greater than 1k and and a goodly number are over 4k.

I don't see how 1k blocks are reasonable on 3390 ecked
devices under any circumstances.

=====
Jim Sibley
RHCT, Implementor of Linux on zSeries

"Computer are useless.They can only give answers." Pablo Picasso

__________________________________
Do you Yahoo!?
Yahoo! Finance Tax Center - File online. File on time.
http://taxes.yahoo.com/filing.html

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions, send email
to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to