Sep 1 ’03
Based on the continued evolution of the z/OS platform, it is only normal that we expect faster processing speeds, more capacity, and price/performance improvements. The z990 announced in May satisfies these expectations and provides new opportunities. I’ll let IBM and others review the announcements for you; I will focus on two z990 opportunities — consolidation and HiperSockets.
Much has been written about IT consolidation that focuses on the server farm. The z990 presents a good opportunity for the consolidation of z900s and 9672s. In the past, the primary obstacle for consolidating mainframes has been the I/O configuration limit of 255 channels. Each step in mainframe evolution has included one or more of the following: faster engines, more engines, faster interconnections, more interconnections, alternative channel types, etc.
However, few evolutionary steps have included changes in the I/O architecture. The z990 is the exception, as it introduces the concept of Logical Channel Subsystems (LCSSes), which will be available in Oct. 2003.
LCSSes will provide a new level of virtualization in the I/O subsystem. The maximum number of channels for an LCSS is 256, which is the same amount for a z/OS, z/VM, or Linux operating system image. This has not changed. What has changed is that the LPARs do not all have to use the same 256 physical channels. On z990, the LCSS is defined first, and describes which physical channels are in which LCSS. LPARs are defined to a specific LCSS rather than to the physical machine.
LCSSes can share channels between LPARs, and a new capability allows LCSSes to span channels, thus allowing channels to be shared between LCSSes. Figure 1 shows two LCSSes and 30 LPARs, as will be available in Oct. 2003. Note that CHPID FF spans across both LCSSes.
The two LCSSes help greatly when consolidating machines. One machine can be migrated to a new z990 using LCSS 0 while maintaining the 256-channel I/O configuration from the original machine. A second machine can be migrated to the same z990 using LCSS 1. The second machine can then continue to use its same I/O configuration with the 256 channels of LCSS 1. Previously, the two I/O configurations would need to be merged in order to share the 256 channels of an earlier generation mainframe. Merging two I/O configurations can be a very difficult task.
HiperSockets were introduced with z/OS V1R2 in October 2001. Before HiperSockets, LPAR-to-LPAR communication was possible with shared Open Systems Adapters. HiperSockets provided an IP network at memory speeds without physical adapters, often referred to as a Virtual Local Area Network (VLAN). The z900 is limited to four VLANs, but the z990 expands this limit to 16. Channel spanning can be used with HiperSockets. This allows multiple LPARs with different LCSSes access to a single HiperSockets channel.
VLANs can be used for LPAR-to- LPAR communication and also with z/VM guests and z990 Integrated Facility for Linux (IFL). Linux applications can access data stored on z/OS at memory speed without any network latency. This reduces complexity and greatly simplifies management of mixed operating systems that can be run on a single z990 or Parallel Sysplex.
WLC SAVINGS WITH CONSOLIDATION
Software pricing can be an obstacle to consolidation. IBM’s Workload License Charges (WLCs) should help with the IBM portion of the software bill. Recall that variable WLCs are based on the Maximum Simultaneous Four-Hour Rolling Average (S4HRA) [see my article in the Feb./March 2003 issue of z/Journal for additional information]. It is unlikely that both machines would have the exact same S4HRA. The result is that as you consolidate machines, the S4HRA decreases. The more “offset” the workloads are, the greater the potential savings.
Consider the following simplified example with two z900 machines:
- A 2064-110 with three LPARs, and the S4HRA for z/OS, CICS, and DB2 is 300 MSUs on April 15 at 1 p.m.
- A 2064-115 with four LPARs, and the peak S4HRA for z/OS and DB2 is 412 MSUs on April 18 at 3 p.m. IMS is also on the machine, and its S4HRA is 215 MSUs on April 4 at 11 a.m.
The Variable WLC software charges are $253,109 as shown in Figure 2.
When these two workloads are moved to a single z990, the new S4HRA is only 650 MSUs, less than the 712 MSUs shown in Figure 2 when the machines were separate. The Variable WLC charge is now $245,297 — for April 2003, the savings is $7,812. How much would the savings be for a year? That depends on how the S4HRA changes month to month.
The effect of combining multiple LPARs greatly lowers the S4HRA. The WLC example applies equally well if you have the opportunity to combine small z900s or 9672s onto the larger z900 machines. Any flat WLC products that are on the two machines would also only be charged for one machine, thus providing further savings (see Figure 3).
The z990 has the processing capability, I/O bandwidth, and networking flexibility to provide opportunities for consolidation and application deployment. For the IBM portion of your software charges, Variable WLC should provide savings. However, you need to work with your other software vendors regarding their pricing strategies. These examples have shown that the z990 and the accompanying new features and functions do provide opportunities and have the potential to reduce your software charges. Z