You can view or modify any read-write Unionfs configfs variables similarly
to the 'timeout' variable above.
+CONFIGURING THE ODF PARTITION
+=============================
+
+The following provides a rough estimate of the ODF partition size and
+configuration. Since regular files in the union appear as zero-length
+files in the odf, they do not utilize any disk blocks on /odf; they do
+consume inodes though. Directories contribute to the disk block usage,
+both from their appearance in the /odf/ns as well as their cached contents
+in /odf/ic. On our setup, where the lower branch contained a Linux kernel
+source tree, we found that majority of the directories consumed less than
+4K disk blocks, and many of those, less than 1K blocks.
+
+Our aim is to minimize the disk block usage per directory, and maximize
+the inodes from the partition. To achieve this, we configure the /odf
+with blocksize of 1K, and create one inode for every 1K bytes.
+
+The following formula should help us arrive at an estimate:
+
+Odf Size = Sum of (2XY/B) over all lower branches, where
+
+X = Size of the lower branch
+ % inodes used on lower branch
+Y = ------------------------------
+ % blocks used on lower branch
+B = Number of 1K blocks-per-inode on the lower branch
+
+
+We can then use the following to format the file system on /odf:
+
+ mkfs.ext2 <odf_device> -i 1024 -b 1024
+
+
+Example:
+
+Consider a single lower partition union:
+
+bash> df /n/lower/
+Filesystem 1K-blocks Used Available Use% Mounted on
+/dev/hdb1 8256604 5681184 2156008 73% /n/lower
+
+bash> df -i /n/lower/
+Filesystem Inodes IUsed IFree IUse% Mounted on
+/dev/hdb1 1048576 43264 1005312 5% /n/lower
+
+
+Here,
+X = 8256604K
+Y = 5/73
+B = 8
+
+odf Size = 2 * 8256604K * 5/(73 * 8)
+ = 141380K
+ = 138M
+
+An /odf partition of size 138M, formatted for 1K blocksize and 1K
+bytes-per-inode should suffice for a union on a 8G lower partition.
+
+
+Caveats:
+
+1. This estimate does not include future branch management which could
+occur on the union. If you would likely be adding more branches to the
+union at a later time, then those requirements will also have to be
+addressed at the time of configuring /odf.
+
+2. Although in principle /odf can be formatted with any file system,
+the estimate above is made keeping ext2 in mind. The motivation behind
+the estimate will remain the same, and can be extended to any other
+file system.
+
+
For more information, see <http://unionfs.filesystems.org/>.