We support one large HSM filesystem, and have recently exceeded
1 million inodes in use (HSM client v3.1.06L, AIX4.2.1). Prior to
this, dsmreconcile took approx 4-5 hours to complete. Since reaching
this threshold, that window dramatically increased ( >14 hrs).

I am aware of APAR IX8836 detailing this problem. Has anyone
gotten a workaround or a fixtest to help deal with this performance
issue since no PTF is yet available? (I've checked the fixtest
directory on index.storsys.ibm.com, but see nothing related.)

This is the third very serious problem we have run into with HSM in the past
couple of months, related to HSM and large filesystems. Presently, we are
running with a fixtest version of the client allowing us to migrate >2TB
of data, as well as running with a patched dsmreconcile binary
allowing us to use more than one data segment.

How are other large HSM sites managing to balance their production
requirements given these weaknesses and seemingly immature nature of
this client?

Thanks.

--Ruth
  Cornell Theory Center
  Cornell Theory Center