Reference result

Edit online

Running the combined workload set in an unconstrained z/VM® environment was called the reference execution. Unconstrained means that an abundance of z/VM real memory was available such that no z/VM paging occurs, and that an abundance of z/VM real CPUs was available such that not all of these CPUs were loaded to their limit. We determined that 512 GiB real memory and 25 real CPUs satisfied these prerequisites.

The reference result is the result of the reference execution.

Table 1 lists the average CPU load and the average memory consumption for the individual workloads that was observed when performing the reference execution.
Table 1. Reference result: Average CPU and memory consumption of the combined workload set

Table with six columns listing a sequence number, the workload, its level, the component or variant, the defined and utilized CPUs and the defined and instantiated memory.

# Workload Level Component or variant CPU [# proc] Memory [MiB]
defined CPU load defined instantiated
1 Database BI 4 2.82 307,200 83,968
2 Transactional WAS medium IHS 1 0.58 700 560
3 WAS 2 1.18 4,096 2,232
4 DB2® 1 0.92 2,048 2,048
5 high IHS 1 0.86 700 575
6 WAS 4 1.72 4,096 2,222
7 DB2 2 1.27 2,048 2,048
8 File system medium page cached 2 0.92 16,384 12,288
9 direct I/O 2 0.88 16,384 2,543
10 high page cached 4 1.60 16,384 16,384
11 direct I/O 4 1.22 16,384 2,572
12 Java™ medium 1 0.92 4,096 3,626
13 high 2 1.82 4,096 3,579
14 Network medium client 1 0.79 1,024 467
15 server 1 0.94 1,024 436
16 high client 2 1.49 1,024 465
17 server 2 1.65 1,024 443
Sum: 36 21.57 398,712 136,456

Observations:

No individual workload (or workload component) used all of its assigned CPU power in terms of virtual processors and processor share.

Likewise, most of the workloads did not make use of all of their virtual memory. However, some workload components, such as DB2 and the high level file system workload using the page cache, made use of all their virtual memory.

Apparently the workload level has almost no impact on the instantiated memory of the guests, except for the file system I/O workload with page cache. This is expected because here the variation in the workload level is related with a change in the amount of files in use.

The aggregated amount of z/VM processing power used by the combined workload set running in an unconstrained environment was about 21.6 CPUs, and the aggregated amount of required real memory – which in this unconstrained environment is identical with the aggregated amount of instantiated memory – was about 133 GiB (1 GiB = 1024 MiB).

Conclusions:

The amounts of real resources needed are significantly smaller than the accumulated amounts of virtual resources configured for the virtual systems. This confirms the statement made earlier that the more or less arbitrarily configured sizes are not a good basis for determining resource overcommitment (see Memory overcommitment factor). Instead, in this paper we define resource overcommitment with respect to the reference result.