11.15 Capacity planning summary

< Day Day Up >

Capacity planning requires measurement, analysis, and knowledge based on experience. This chapter has shown some of the more relevant areas to measure. Using these techniques, you have the ability to measure your existing platforms and project your requirements to a z/VM environment.

For processor utilization, you have the ability to measure the existing workload on Intel or other platforms, and project the requirement on a zSeries processor.

For storage (RAM, in Linux terms), there was about a 400 megabyte requirement for each Linux server, including storage and virtual disk for the workload tested. Early "by the book" measurements showed a requirement of 500 MB to support a small workload (300 users). By reducing the storage sizes, one run showed a total of 360 MB requirement to run a large workload (1500 users).

Storage requirements will vary significantly. In z/VM, there is the option of varying storage based on workload. By using virtual disks as swap, very little resource is consumed by the virtual disks, but when the workload requires more storage, it is available. Because the virtual disks reside in real storage when active, they are extremely fast.

Virtual disks in z/VM provide a completely different aspect of performance for both the Linux-knowledgeable and the VM--knowledgeable. For example, Linux administrators would never consider swapping at 1000 swap pages per second. However, one of the experiments we ran did this with no ill effect on end user response times.

< Day Day Up >