In this issue, I’ll cover the following:
1. Latest CMOS Processors
2. How TUNING Letter Subscribers Can Obtain the New MIPS Ratings
3. Correction to RESMIL Recommendation
1. Latest CMOS Processors
On May 7th IBM announced their new G5 series of processors, the 9672-Rx6 series (announcement #198-115). This series consists of sixteen new CMOS processors, most based on a 115 MIPS uni-processor with one turbo-charged model that comes in at 900 MIPS.
The effective speed and capacity of the processors confirms that CMOS can now compete with all of the current bipolars and with all but one of the current HDS Skyline models.
HDS followed a week later on May 14th and announced their comparable Pilot P8 series plus two additional turbo models. Amdahl is expected to announce their next series of Millennium CMOS processors that will surpass the speed and capacity of these latest IBM and HDS models, but with an availability date of 1Q99.
Other announcements, such as extensions to parallel sysplex that are provided by both vendors and IBM’s latest software pricing announcements, are fully discussed in our Cheryl Watson’s TUNING Letter, 1998, No. 2, which will be mailed in late May.
Neither vendor announced performance estimates. The Processor group and MSUs listed below are our best estimates at this time.
The new CMOS models are:
The 9672-R06 and Pilot P08 are currently announced as coupling facility models that can be upgraded to standard S/390 processors.
2. How TUNING Letter Subscribers Can Obtain the New MIPS Ratings
A major benefit of subscribing to Cheryl Watson’s TUNING Letter is receiving, at no extra charge, our extensive CPU Chart and its updates. The CPU Chart contains CPU models from all three major vendors, the number of CPUs, average, min, and max MIPS, MIPS per CPU, MIPS by LSPR workloads, processor groups, MSUs, and version codes. It even provides the processor groups in descending order and MSUs for easy evaluation of options when trying to size a new machine.
We’ve update our CPU Chart with the MIPS for these models, and the new chart will be sent out to all subscribers by the end of May. Subscribers to Cheryl Watson’s TUNING Letter may ask to receive a fax or email containing MIPS for the new models by sending an email to admin@watsonwalker.com (be sure to include your company name and the address to which the TUNING Letter is mailed).
3. Correction to RESMIL Recommendation
Steven Niebauer, of Wachovia Corp., brought to our attention an error in our recommendation of the RESMIL parameter of the GRSCNFxx parmlib member (see page 29 in Tuning Letter 1998, No. 1). In this article we accidentally specified that the RESMIL units were seconds rather than milliseconds. Steven also suggested we correct and clarify our definition of the RESMIL parameter. RESMIL specifies the minimum amount of time that a system must hold the RSA-message before passing it along. For your convenience, we are providing here the new complete write-up. Thanks Steven.
For those of you that are not TUNING Letter subscribers, our 1998 No.1 TUNING Letter issue contained the first part of a two-part article that identifies most of the changes that might be needed in SYS1.PARMLIB and similar parameter libraries whenever you make a software or hardware configuration change. The RESMIL parameter was discussed in this article during a review of SYS1.PARMLIB member GRSCNFxx.
_____ü RESMIL(n)
Condition: Adding or removing systems from a sysplex. Adding or removing systems in a multi-system GRS configuration.
The RESMIL parameter defines, in milliseconds, the RSA-message residency time. The default of 10 milliseconds will work for most installations, but may cause some performance problems.
An RSA-message contains the information about requests for global resources, and it is this message that is passed around to each system in the ring. When an RSA-message is received on the system it is processed by GRS. The RESMIL parameter defines the minimum number of milliseconds a system must hold an RSA-message before passing it along. It does not affect the time a message is processed, but rather the minimum time a message will be resident on a system. A RESMIL(0) would indicates that the system tunes the residency time in a range with a minimum of zero. A RESMIL(10) would indicate that the system maintains a residency time of at least 10 milliseconds.
Impact: MEDIUM A large RESMIL time in a small LPAR (or on a slow machine) in the SYSPLEX could cause other systems to wait long for a response to a request. This could cause higher response times in the other systems. A typical situation occurs if you have just added a small LPAR to a sysplex and that LPAR is only active 5% of the time. It may take a long time for the LPAR to release the RSA-message. Most sites are finding that the ability to specify RESMIL(0) or RESMIL(5) provides significant performance benefits. IBM also recommends that you specify the same RESMIL values on all your systems in the GRS ring regardless of the varying size of the processors in the ring.
That’s all for now. Stay tuned!