Our top recommendation for improving DFSMShsm throughput is to use RLS for the 
HSM CDSes.  In nonRLS mode, each HSM locks access to the CDSes while it 
performs its queued up I/O requests.  While that host has the CDSes locked, all 
other HSM hosts are queueing up their I/O.  When a host gets exclusive access, 
it has to first flush all of its buffers because they are no longer valid and 
then do a VSAM VERIFY.  All of the buffers are then repopulated for the I/Os 
that have been queued up.

All of that overhead goes away with RLS.  Also, Read I/Os are faster, so read 
intensive workloads like ExpireBV, Secondary Space Management and Audit perform 
faster, even when there isn't alot of multiple host concurrent workload.  The 
higher the amount of concurrent HSM host activity that you have, the greater 
the improvement that you'll see with RLS.

In z/OS V2R1, DFSMShsm enhanced its error recovery for SMSVSAM server errors.  
Prior to V2R1, when there was an SMSVSAM server error, HSM lost access to the 
CDSes, so HSM would just take a fatal abend.  (Yes, very ugly, we know).  
Starting in V2R1, DFSMShsm will hold I/O when it identifies that there is an 
SMSVSAM error.  HSM waits for the SMSVSAM server to restart.  After the server 
restart, HSM will automatically retry all of the I/Os that were held.  The only 
I/Os that have to be failed are Update I/Os after a Read for Update, when the 
Read for Update was before the server error.

Glenn Wilcock
DFSMShsm Architect

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to