I'm still confused ... here's my interpretation, based on previous postings,
IBM manuals, and some googling:
CPU% is time spent "on behalf of" the address space (R791TCPU)
It includes all the TCB (ASCBEJST) and SRB (ASCBSRBT) time for the address
space's own internal work
It also includes any enclave time (ASSBASST) spent in another (server) address
space (e.g., DB2)
Therefore for a "client" address space like DDF it should represent all
DDF-related work
For a "server" address space like DB2 it should represent only the DB2 internal
(overhead) work
ECPU% is time spent "in" the address space (R791TCPC)
It also includes all the TCB (ASCBEJST) and SRB (ASCBSRBT) time for the address
space's own internal work
It does not include any enclave time spent in another (server) address space
It does include enclave time (ASSBPHTM) that this (server) address space
executes "on behalf of" a client
Therefore for a "server" address space like DB2 it should represent all
DB2-related work
For a "client" address space like DDF it should represent only the DDF internal
(overhead) work
Of course, some address spaces might both create and service enclave work, but
hopefully that is not an issue in my case.
Now here's a concrete example ... (I changed a few names for clarity purposes
only).
JOBNAME CPU% CPU-Time SCPU% Workload SrvClass ECPU-Time ECPU%
DB2XDIST 1.50 4,403.60 85.00 DDF DDF 408,886.43
36.96 DB2QDIST 16.13 160,199.54 85.00 DDF DDF 508,487.94
18.40 PCICSA04 40.36 467,627.88 85.00 STC CICSSTC 467,627.88
17.75 DB2XDBM1 9.09 33,046.14 85.00 DB2 DB2X 33,046.14
4.00 OMEGDB2 4.02 30,483.05 85.00 STC MONITOR
30,483.05 1.77 TCPIP 1.49 26,865.07 85.00 SYSTEM SYSSTC
26,865.07 0.65 DB2QDBM1 1.44 47,296.32 85.00 DB2
DB2Q 47,296.32 0.63
DB2XDIST is a PeopleSoft application. Although there are some batch jobs,
most work comes from "off the box".
DB2QDIST also comes from off the box but is an in-house application with a
different profile -- it executes a high volume of very tiny requests, hence I
anticipate much higher enclave management overhead.
By my logic, I expected to see ECPU% >> CPU% for the DB2 workloads (just like
in the previous posting) because ECPU includes the enclave CPU executed
"within" the DB2 address space "on behalf of" the DDF address space.
Similarly I expected to see CPU% >> ECPU% for the DDF workloads (Amy and David
in the previous posting) because CPU includes that same enclave CPU.
Finally, I would expect to see ECPU% = CPU% for non-enclave workloads like CICS
(John in the previous posting -- we're not even running threadsafe in that
region). Note that the total CPU-Time and ECPU-Time do match perfectly.
Clearly, I must have misunderstood something . . . can anyone put me straight?
It looks to me almost as though the enclave CPU for requests originating "off
the box" is simply not reported in the CPU column, but only in the ECPU column.
In actual fact, when I re-calculate the values by pro-rating all the CPU%
fields down so that my non-Enclave address spaces give the same value in both
CPU% and ECPU%, the difference between the ECPU and the re-calculated CPU
(presumably representing the actual NP-SRB/Enclave time) seems entirely
appropriate. Further, I have verified that the interval CPU-Time for the CICS
region is identical to the interval ECPU-TIME. Therefore the CPU% and ECPU%
should be identical unless something is not being reported. This is
definitely not what is suggested by the SDSF help definition of these fields or
by any other documents I've found so far.
If my calculation is correct, I would suggest that the correct interpretation
of the fields on my SDSF display (supposedly "vanilla") are:
CPU-Time is the CPU Time (TCB + SRB) for the address space, EXCLUDING any
NP-SRB/Enclave time
ECPU-Time is the CPU Time (TCB + SRB) for the address space, PLUS any
NP-SRB/Enclave time [i.e. CPU used "on behalf of" this address space -- the
definition of CPU-Time]
The "Server" CPU Time (TCB + SRB + served NPE-SRB/Enclave time -- CPU used
"within" this address space -- the definition of ECPU-Time) is not available on
my display; the NP-SRB/Enclave time is also not directly available.
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
