You can use the RMF Postprocessor to generate long-term
overview reports for your Spark workload. Unlike the
interactive Monitor III sessions, Postprocessor reports allow you to analyze your Spark workload performance over the
long term and fine tune its resource assignments. For more information about running the
Postprocessor and other report options, see z/OS RMF User's Guide.
Workload Activity report
The Workload Activity report (WLMGL) presents a
summarized view of a WLM-managed workload. You can select to view this report for a specific service
class, report class, WLM workload, and various other types. For instance, if you place all of your
Spark processes in a single
service class, you can generate a Workload Activity report that tells you the actual amount of
memory and CPU consumed by that workload. Figure 1
shows an example of the Workload Activity report for the ODASSC1 service
class.
Table 1 describes some of the fields in
the Workload Activity report that are more relevant to Spark.
Table 1. Selected fields in the RMF Workload Activity
report
Field heading
Entries
SERVICE TIME
CPU
Time, in seconds, consumed on general-purpose and special-purpose processors.
IIT
I/O interrupt time, in seconds.
IIP
zIIP service time, in seconds.
APPL %
CP
Percentage of the processor time used by transactions running on general-purpose
processors.
IIPCP
Percentage of the processor time used by zIIP-eligible transactions running on general-purpose
processors. This is a subset of the APPL % CP value.
IIP
Percentage of the processor time used by transactions executed on zIIPs.
STORAGE
AVG
Weighted average number of central storage frames allocated to active ASIDs.
TOTAL
Total number of central storage frames allocated to resident ASIDs.
PAGE-IN RATES
SINGLE
Average rate at which pages are read into central storage while transactions are resident in
central storage.
BLOCK
Rate of demand page-ins from DASD for blocked pages.