This report shows the CPU usage during one RMF interval, broken down by CPU type, so it can be used for decisions about the use of special processors.
The fields in the report are as follows:
Logical processor number in the LPAR.
Can be either CP for general processor or AAP/IIP for zAAP and zIIP respectively.
Percent of time the processor was online during the interval. It can be less than the interval if the processor is running under WLM and was reconfigured during the interval.
The LPAR busy time is defined as follows:
The BUSY and WAIT PCT is the percentage of online time during which the processor was busy (working) or waiting (idle).
The MVS busy time is defined as follows:
Without Hiperdispatch, PARKED TIME is always 0. (See the explanation of parked time below.) The BUSY PCT and WAIT PCT is the percentage of non-parked time (online time reduced by parked time) during which the processor was busy (working) or waiting (idle).
Percentage of total time during which the processor was parked. Only processors with a "LOW" amount of physical processor share on LPARs with HiperDispatch enabled can be parked. (See the explanation of parked time below.)
Reported CPU serial number.
What HiperDispatch does is manage the mapping of logical CPUs to physical CPUs on z10 processors. This optimizes the use of processor cache by localizing work. Redispatching work on the same physical processor maximizes the probability that the work's working set will still be in cache.
z/OS is no longer unaware of running under PR/SM. z/OS and PR/SM cooperate to localize work to a core, a chip, or a book. Each processor (core) on a chip has local L1 and L1.5 cache. L2 cache is shared between the processors on a chip. Physical memory (up to 384 GB) is shared between the chips on a book. Each level of memory (L1, L1.5, L2, memory on this book, and memory on other books) is dramatically slower than the next lower layer.
HiperDispatch optimizes this hierarchy by packing work within an LPAR onto as few processors as possible. It does this by assigning processor weight to logical processors. The PR/SM weight assigned to a logical processor is called "polarity." A logical processor with a polarity of 100 is dedicated, with polarity zero called "parked." That is, the logical processor is assigned, online, but gets no work. That is how "parked" time is reported in RMF.
A large (multibook) configuration will get the most out of HiperDispatch. IBM claims up to 10% ITR. A single-book, lightly loaded configuration will not profit, and may lose ground. For more information, see the paper by Don Deese, which is available for download from www.cpexpert.com and will be presented at CMG in 2008.
1 4/11/yy PAGE 1 CPU ACTIVITY REPORT SYSTEM ID: CA31 DATE: yy/03/05 TIME: 22:59:00 INTERVAL: 00:14:59 - CPU CPU ONLINE LPAR BUSY BUSY WAIT MVS BUSY BUSY WAIT PARKED CPU NO. TYPE TIME PCT TIME PCT PCT TIME PCT PCT TIME PCT SERIAL 0 4 AAP 100.00 00:00:14.97 1.66 98.33 00:00:14.97 1.66 98.33 0.00 0CE000 TOTAL/AVG 00:00:14.97 1.66 98.33 00:00:14.97 1.66 98.33 0.00 0 0 CP 100.00 00:07:42.32 51.37 48.62 00:08:43.63 58.18 41.81 0.00 0CE000 1 100.00 00:07:36.83 50.76 49.23 00:08:36.03 57.33 42.66 0.00 0CE000 2 100.00 00:07:24.24 49.36 50.63 00:02:07.17 24.28 75.71 41.81 0CE000 3 100.00 00:07:10.75 47.86 52.13 00:00:00.00 100.00 0.00 100.00 0CE000 TOTAL/AVG 00:29:54.16 49.83 50.16 00:19:26.84 50.21 49.78 35.45 0 5 IIP 100.00 00:00:15.46 1.71 98.28 00:00:15.40 1.71 98.28 0.00 0CE000 6 100.00 00:00:13.22 1.46 98.53 00:00:13.16 1.46 98.53 0.00 0CE000 7 100.00 00:00:14.13 1.57 98.42 00:03:21.89 36.64 63.35 38.78 0CE000 TOTAL/AVG 00:00:42.81 1.58 98.41 00:03:50.45 9.80 90.19 12.92
|
Copyright © 2012 CA.
All rights reserved.
|
|