1. TUNING Letter 2003, No. 5 Updates
The forty-two page 2003, No. 5 TUNING Letter was emailed to electronic subscribers on January 9. The print issues will be mailed this week. There are three corrections we would like to bring to your attention. These will be corrected in the printed issues and Linda May will be sending out corrected PDF files to electronic subscribers within the next few days.
Incorrect Session Number
Jack Schudel from University of Florida pointed out that my WLM SHARE session was listed as session 2939, but it should be 2539. Oops! Although session 2939 is a real session that will teach you about UNIX file system availability, we really hope you are planning to attend 2539 so that you can hear some of our exciting announcements.
IRD Vary CPU Management
Nils Abrahamsson from Volvo IT and Al Sherkow of I/S Management Strategies, Ltd. pointed out some errors in our article on IRD. The Vary CPU Management section incorrectly stated that you could specify the initial, minimum and maximum number of LPs for an LPAR. Actually, you can specify only the initial and reserved number of LPs. The following section should replace the section on pages 22-23 of the article:
IRD Vary CPU Management
If you do nothing after you install z/OS but turn on IRD CPU weight management, you are ready to use IRD Vary CPU Management. Because CPU weight management is on by default, to turn it off you must add a VARYCPU=NO statement to the IEAOPTxx member for each image. IRD Vary CPU Management is active when either the VARYCPU statement is missing or you specify VARYCPU=YES and you have activated IRD CPU weight management on the HMC.
If both of these conditions are satisfied, then IRD will dynamically control the number of LPs assigned to each LPAR in an LPAR cluster. The HMC consoles, as shown in Figure 2, are used to define the initial and reserved number of LPs for an LPAR. When you IPL, the number of LPs is set to the initial value. WLM can change the number in order to keep the number of LPs to a minimum (to reduce unnecessary overhead), but it cannot exceed the initial value. This varying of LPs can occur several times per interval. In fact, for most installations we’ve studied, the number of LPs online at the end of almost every 15-minute interval was different. In one example, the number of LPs varied from 7 to 12 (the maximum) during prime shift. The ‘initial’ value is just that – the IPL value and the value that indicates the number of shared CPs that are available for the LPAR. If you manually configure online more LPs to an LPAR, WLM and IRD can then use those additional LPs.
The reserved number of LPs refers to LPs that you can manually configure online, either adding them from the current pool of available CPs or after dynamically adding them with CoD (Capacity on Demand) or OOCoD (On/Off Capacity on Demand).
Although the number of LPs will start at the initial value and will be changed by IRD for performance considerations, they may also be changed when the weight is changed. After all, there’s no point in increasing the weight if you don’t have enough LPs to use the additional capacity. Similarly, you might see a reduction in the number of LPs if the weight is decreased. The number varied online by WLM/IRD cannot exceed the value you specified as ‘initial’ plus the number you manually configured online.
Although the manual says that IRD tries to keep the LPs to a minimum, that’s not always true. At high utilizations, the LPARs will tend to have the minimum number of LPs in order to meet the goals. At low utilizations, however, IRD will allow the number of LPs to increase (normally two more than are really needed) in order to have available LPs when a burst of work appears on the system. Because of this behavior, on lightly loaded systems IRD may actually increase the amount of LPAR overhead in an installation. This can affect chargeback, but should not cause performance problems.
For production LPARs, IBM recommends that you set the initial number of LPs equal to the number currently available on the machine so that WLM and IRD will have full control of the system. If the system is not busy, most of the LPs will be inactivated. You should set the reserved number of LPs equal to the maximum number of CPs that could be added to this machine (this will allow for future upgrades). We recommend that you limit the initial LPs on your test LPARs and smaller LPARs. You don’t want to allow WLM and IRD to assign too many LPs to them if they happen to have a very large piece of work.
The initial number of LPs is a very important consideration. The service units per second are set at IPL time and are not changed when IRD changes the number of LPs. However, if you manually configure an additional CP online, then the service units per second are changed. So manually changing the number of LPs will affect the service units per second calcaulation, but that will not be true for IRD. If you are doing charge back using service units, your charges may become more variable and unrepeatable when you start running IRD.
Thanks to Bob Richards from SunTrust Banks for catching a typo on page 26. The phrase at the end of the first paragraph should read “when SYSA was increased, SYSC was decreased.”