We recently had a problem restoring from a 5.1.7.2 server under z/OS to
a 5.1.5.0 client under HP-UX. We had to restore about 3.5 million files.
The restore ended when a disk failure knocked out the file system containing
the mount point for the file system we were restoring. We had restored
about 3.0 million files at that point. The restore was supposed to be
restartable, but was terminated completely when the I/O error occurred.
We started a new restore with the -ifnewer option, which we hoped would
be a relatively cheap way of skipping over the files already restored.
The new restore performed very poorly, and we eventually discovered that
a 'show memu SHORT' command (a memory management diagnostic peculiar to the
z/OS implementation of TSM) reported FreeHeld larger than MaxQuickFree.
At first we thought this was the cumulative result of multiple restores
and other server activity. We suspended the restore, bounced the TSM
server to clean up memory management, and restarted the restore. Even
with no other significant activity on the server we saw FreeHeld climbing
steadily and exceeding MaxQuickFree about an hour after restarting the
restore. At that point we gave up on the -ifnewer option and restored
all 3.5 million files. Is the apparent memory leak a known problem with
the -ifnewer option?