Storing and Managing Data in Today’s z/OS
Environment
Michael E. Friske
Storage Challenges
Compounded growth Cost reduction pressure Changing regulatory and business
requirements Higher availability requirements Data life cycle management
Hardware Solutions That Can Help Address These
Challenges Storage Tiering
– IBM DS8870 with EasyTier– EMC VMAX with FAST– HDS VSP with HDS Dynamic Tiering
Virtual Tape– IBM DS7700– EMC DLm
– Oracle VSM– CA-Vtape
(software/hardware)
Traditional Storage Hierarchy
High Performance
Low Cost
Cache or SSD
Enterprise Disk
ML1 Disk
ML2 Tape
Data Life Cycle
Active data– Stored on primary disk
Less active data– Migrate to ML1 disk– Uses MIP’s– Must be recalled to use
Inactive data– Migrate to ML2 tape– Uses MIP’s– Must recall to use
Primary Disk
ML1Disk
ML2Tape
SMS MGMTCLAS
Number of days on primary disk Number of days on ML1 volumes When to expire
Management criteria– Number of days since the create date– Number of days since the last reference
date
Disk Options Flash Drives (100GB, 200GB, 400GB,
1.6TB, and 3.2TB) Solid State Drives (400GB) SAS Drives
– 10,000 RPM (300GB, 600GB, 900GB, and 1.2TB
– 15,000 RPM (146GB and 300GB) Nearline SAS Drives
– 7,200 RPM (1TB, 2TB, 3TB, and 4TB)
Determining What Data Went on Which Drives
Disk subsystems with a mix of different types of disk drives– Analyze RMF and/or SMF reports– Map SMS storage groups to specific disk
drives– Monitor for changes in access patterns
A homogenous disk subsystem that works for all types of data
Hardware Based Tier Management
IBM DS8870 EMC VMAX HDS VSP
Easy Tier Fully Automated Storage Tiering (FAST)
HDS Dynamic Tiering
First Generation Tiering
The Disk subsystem monitored the I/O activity for each extent
Extents were dynamically relocated to different tiers based on internal algorithms
Shortcomings of First Generation Tiering
Data is moved at the extent level, not on a data set boundary
Data sets move around due to reorgs, defrags, copying, etc.
Patterns of access change based on the time of day or the day of the week
Data is placed on a storage tier based on access, not SLA’s
Addressing the Shortcomings
Hints from the host– z/OS 2.1 Class Transitions– Hitachi Tiered Storage Manager
MGMTCLAS - Expiration
MGMTCLAS - Migration
MGMTCLAS – Class Transition
MGMTCLAS – Transition Technique
Easy Tier with Class Transition
Enterprise$
Nearline½ $
Class transition after 120 days
Migrate after 366 days
Recall
Transition Copy Technique
FRP - FR PREFERRED FRR - FR REQUIRED STD - STANDARD PMP - FC PRESMIRPREF PMR - FC PRESMIRREQ
Serialization Error Exit
DB2 - Invoke DB2 to close and unallocate the object. If this is successful, the object is serialized and moved and DB2 is invoked to reopen the object.
CICS - Invoke CICS to take the object offline. If this is successful, the object is serialized and moved and CICS is invoked to reopen the object.
ZFS - Invoke zFS to unmount the data set. If this is successful, the data set is serialized and moved and zFS is invoked to remount the data set.
EXIT - Enables users / ISVs to provide an exit that will be invoked before and after transitioning an allocated data set. The data set is only transitioned if serialization is obtained after the first invocation of the exit.
NONE
Transition Invokes the ACS Routines
Once DFSMShsm determines that a data set has met the Class Transition criteria specified by the Management Class, it invokes the ACS routines to determine what the transition should be– ACS Routines are invoked with new ACS environment
(&ACSENVIR) of SPMGCLTR, for 'space management class transition'
– The following routines are invoked (in this order)» Storage Class» Management Class» Storage Group
– Any or all can be transitioned If the classes and storage group returned match the existing
classes and storage group, then no transition occurs
STORCLAS Routine
IF &ACSENVIR = 'SPMGCLTR' THEN SELECT (&STORCLAS)
WHEN (‘FLASH') SET &STORCLAS = 'EASYTIER'OTHERWISE SET &STORCLAS = &STORCLAS
ENDELSE ...IBM
MGMTCLAS Routine
IF &ACSENVIR = 'SPMGCLTR' THEN SELECT (&MGMTCLAS) WHEN ('NOML2') SET &MGMTCLAS = 'ML2OK' WHEN ('DB2NEW') SET &MGMTCLAS = 'DB2AGED' OTHERWISE SET &MGMTCLAS = &MGMTCLAS ENDELSE...IBM
STORGRP Routine
IF &ACSENVIR = 'SPMGCLTR' THEN SELECT (&STORCLAS) WHEN (‘FLASH') SET &STORGRP = 'TIER0' WHEN ('EASYTIER') SET &STORGRP = 'TIER1' OTHERWISE SET &STORGRP = &STORGRP ENDELSE ...IBM
Example – New Data Set Initial access
– Random reads & writes with periods of high access
STORGRP = TIER0– Flash– Enterprise
MGMTCLAS– No migration– Transition after 60
days
Flash
Nearline
Enterprise
TIER1
TIER0
Example – Access Drops Off
Access after 3 months– Accessed less– No write activity
STORGRP = TIER1– Enterprise– Nearline
MGMTCLAS– Migrate after 30
days of inactivity– Expire after 7 years
Flash
Nearline
Enterprise
TIER1
TIER0
Transition after 60 days
Example - Inactive
Flash
Nearline
Enterprise
TIER1
TIER0 Virtual Tape
Recall
Migrate
Cost Savings
Reduce CPU costs by eliminating ML1– No need to compress or decompress data– Eliminate recalls from ML1
Reduce CPU costs with Class Transition– May be able to use FlashCopy– May be able to wait longer before migrating to
ML2 Reduce the purchase price for new disk
subsystems
Virtual Tape Subsystems
IBM TS7720 EMC DLm Oracle VSM
Cost Reduction with Virtual Tape
Reduces floor space requirements Reduces the need for tape handlers Replicate tape data to remote site for
disaster recovery Batch cycle runs faster Tapeless tape subsystems
– Eliminate the need to tape handlers– No need to buy physical tape media
zEnterprise Data Compression
zEDC Express is an IO adapter that does high performance industry standard compression
Used by z/OS Operating System components, IBM Middleware, and ISV product
Required z/OS 2.1 and EC12 or BC12
SMS Compression with zEDC
Can be used in place of generic or tailored compression for QSAM and BSAM data sets
CPU savings for existing users of SMS compression
Disk savings for users who are not currently using SMS compression
Other Features That Can Reduce HSM CPU
Consuption Use RLS to access the HSM CDS’s Replace Interval Migration with On-Demand
Migration Take advantage of Multiple Address Space HSM
(MASH) support Implement the Common Recall Queue Implement Fast Subsequent Migration Avoid reorganizing the HSM CDS’s too frequently HOLD LOG if log data is not required
RLS for HSM CDS’s
Reduces CPU overhead for HSM due to less calls to GRS, less buffer invalidation, and less I/O to disk
Reduces elapse times for long running tasks like Primary Space Management, Secondary Space Management, EXPIREBV, Recycle, and Audit
HSM On-Demand Migration
Does not check every volume each hour to see if the volume is over the high threshold
Eliminates the spike in CPU usage for HSM at the top of each hour
Will only process a volume once after it goes over the high threshold
HSM MASH
Can be used to run HSM backup and space management functions at a lower priority (different WLM Velocity Goals)
Can be used to spread tasks out to more address spaces and to reduce SYSZTIOT contention
HSM Common Recall Queue
Allows multiple hosts to process recalls Can be used to limit tape processing on certain CPU’s Helps with workload balancing Multiple recalls for different data sets on the same
tape from different CPU’s can be processed on a single CPU
Allows higher priority requests across the sysplex to get process first over lower priority requests
Keeps recall requests if HSM is shut down or goes down unexpectantly
Fast Subsequent Migration
Saves CPU costs associated with re-migrating data sets that were recalled but not updated
Reduces the need to run RECYCLE
HSM CDS Reduction
Turn on CA Reclaim for the HSM CDS’s Do not reorganize the HSM CDS’s just
because they have CI and CA splits Only reorganize the HSM CDS’s to
reclaim space
Other Cost Avoidance Measures
Clean up the errors HSM finds each night
Analyze MGMTCLAS definitions periodically, and adjust them if necessary
Check for data sets in the wrong MGMTCLAS
Data That Can Be Used for Analysis
HSM Logs HSM FSR records DCOLLECT data
Questions ?