Optimizing storage class memory usage

How you improve your use of storage class memory (SCM).

Run the following command:
D XCF,STR,STRNAME=IBM1SCEN1
As the structure was already full with message data, because of the previous tests, part of the rebuild involved pre-staging some of the messages from the structure into SCM. This process was initiated by using the previous command. The output from this command produces, for example:
ACTIVE STRUCTURE
----------------
ALLOCATION TIME: 06/17/2014 09:28:50
CFNAME : CF01
COUPLING FACILITY: 002827.IBM.02.00000000B8D7
PARTITION: 3B CPCID: 00
STORAGE CONFIGURATION ALLOCATED MAXIMUM %
ACTUAL SIZE: 1024 M 1024 M 100
AUGMENTED SPACE: 3 M 142 M 2
STORAGE-CLASS MEMORY: 88 M 4096 M 2
ENTRIES: 120120 1089536 11
ELEMENTS: 240240 15664556 1
SPACE USAGE IN-USE TOTAL %
ENTRIES: 84921 219439 38
ELEMENTS: 2707678 3149050 85
EMCS: 2 282044 0
LOCKS: 1024
SCMHIGHTHRESHOLD : 90
SCMLOWTHRESHOLD : 70
ACTUAL SUBNOTIFYDELAY: 5000
PHYSICAL VERSION: CD5186A0 2BD8B85C
LOGICAL VERSION: CD515C50 CE2ED258
SYSTEM-MANAGED PROCESS LEVEL: 9
XCF GRPNAME : IXCLO053
DISPOSITION : KEEP
ACCESS TIME : NOLIMIT
MAX CONNECTIONS: 32
# CONNECTIONS : 1
CONNECTION NAME ID VERSION  SYSNAME JOBNAME  ASID STATE
--------------- -- -------  ------- -------  ---- ----------------
CSQEIBM1CSQ301  01 00010059 SC61    CSQ3MSTR 0091 ACTIVE
Note the following from the output of the command:

  • That STORAGE_CLASS MEMORY provides confirmation that a MAXIMUM of 4096 MB of SCM has been added to the structure.
  • The ALLOCATED figure for the amount of STORAGE-CLASS MEMORY used for pre-staging. There is now free space in the structure where there was none before SCM was added.
  • The amount of AUGMENTED SPACE used to track SCM usage.
  • The point at which the pre-staging algorithm starts to move data from the structure into SCM is when the structure is 90% full. This is indicated by the non-configurable SCMHIGHTHRESHOLD property.
  • The point below which the prefetching algorithm starts to move data from SCM into the structure is when the structure is 70% full. This is indicated by the non-configurable SCMLOWTHRESHOLD property.

We can now test various ways to optimize the use of SCM. Note the following:

  • After SCM is used to store messages, we cannot alter the structure until you have removed all the data from SCM.

    In this case, that means that the entry-to-element ratio is frozen at the value that was in place when SCM was first used. We must carefully ensure that the structure is in the state we want, before the pre-staging algorithm starts moving data into SCM.

  • Is the current structure size correct before using SCM?

    For example, have you increased INITSIZE from 512 MB to a SIZE of 1 GB?

    If we do not do this, it is possible that although you enabled your structure for auto-alteration, the pre-staging algorithm will start to move data into SCM before the alteration has a chance to start. As a result, the structure is frozen using 512 MB of real storage.

  • Is the entry-to-element ratio correct before using SCM?

    The goal of this scenario is to increase the number of offloaded message pointers that can be stored in the structure and SCM as a whole, as well as keeping as many messages entirely in structure storage as possible. Accessing these messages is faster than accessing messages on SMDS.

    Therefore, we need to have a structure that starts with an entry-to-element ratio that is good for storing messages, and then transitions to a ratio that is good for storing message pointers before the prestage algorithm first starts. This transition can be achieved, in part, by making use of the IBM MQ offload rules.

Change the offload rules by issuing the following command:

ALTER CFSTRUCT(SCEN1) OFFLD1SZ(0K)
We might have to carry out several runs to optimize the entry-to-element ratios. The following table shows possible improvements in the number of messages put on the queue during the different phases of the emergency storage scenario.

Test description Number of messages Time to fill queue (seconds)
Basic configuration 27,850 3.2
SMDS with default offload rules 205,000 158
SCM with default offload rules 828,610 469
SCM with adjusted offload rules 1,135,775 679
The last row in the table shows that adjusting the offload rules had the required effect.

We need to examine the system to see if we can improve on these figures in any way. For example, you might run out of available SMDS storage. If we can allocate more SMDS storage you should be able to increase the number of messages on the queue quite significantly.

Parent topic: Emergency storage - basic configuration