How Busy Is Too Busy? Finding the Right Metrics to Reduce Mainframe Costs
Since the days when processor time was costly (and you input a job on punch cards), the CPU Busy metric has seen intense focus. There are so many ways to look at the metric, all having vastly different meanings. Virtualization made it even more complicated. But for many in the performance monitoring and capacity planning field, this is still a very critical number. But is it the most important number?
Why Does CPU Busy Matter?
Aside from the need to do capacity planning, CPU Busy was of concern to business stakeholders whose IT bills were often factored on CPU usage. Traditionally, the chargeback cost per CPU hour included the all-in cost of the datacenter, because the metric was CPU.
For this reason, the business often challenged IT to reduce the CPU cost of their applications. But, as we know, the cost of the datacenter is less and less attributable to the cost of the processors. Instead, the two most critical factors in chargeback costs are people and software.
Chargeback cost = CPU demand + people cost + infrastructure cost + software cost
Cutting people costs by reducing systems staff would end up hurting you in the long run, as it would require more productivity fewer less people, leading to diminished performance from your team. That’s a path you don’t want to take, so let’s look at software costs.
Consider the Costs
For many years, most of us didn’t really look at this number because we had to have software regardless of its cost. Similar to the logarithmic growth of Microsoft Word over the years, the CPU demand of mainframe software has also increased. We use more and more memory and CPU, but we probably haven’t changed how we use it, that much anyway.
We mostly only look at the CPU cost of applications when we are managing capacity; therefore, we may only recognize the cost of software when we get a usage report from IBM. Or maybe we don’t get that report; perhaps it only goes to the finance department. But be sure of this—that cost does go into the cost of applications and reduces the profitability of the company.
So, what if you looked at software cost even more closely than CPU cost when trying to satisfy the user?
If you are using VWLC (sub-capacity pricing) then you’re in luck. Changes in transaction mix and transaction size as well as new workloads can cause the rolling 4-hour average (R4HA) to increase, leading to big increases in your bill from IBM.
While it can be challenging to do as much as you would like with your OLTP workloads, one area of low-hanging fruit is batch. But who thinks about batch? It’s that set of jobs that you hope is running smoothly behind the scenes while you sleep. If anything, you’d like to think about it even less.
The R4HA includes batch, and when anything goes wrong or your teammate decides to run a huge SAS job or a trace during peak time, you are going to see an impact. Maintenance jobs and database reorgs can also impinge on peak hours, forcing you to pay more when, in fact, your CICS and IMS workloads didn’t change.
If you’re not looking at batch, you’re not saving as much money as you could. But who has time for that?
Make the Investment
Consider investing in a software package that will pay for itself very quickly. Compuware ThruPut Manager can do the work you don’t have time for and manage batch workload to ensure that it won’t cause problems for online performance. ThruPut Manager will also help you reduce your R4HA, key to reducing your costs.
In our webcast, “How Busy Is Too Busy? Automating Your System for Maximum Throughput,” hosted by IBM Systems Magazine, we dove into issues affecting utilization and the key role that ThruPut Manager plays in optimizing your batch workloads to keep your machine running at the performance sweet spot. (The on-demand replay is available–just click on the webcast title above.)
The CPU Busy metric may give you an overall measurement of utilization, but we know that it’s not a linear function. Throwing more and more work at the machine results in diminishing returns and requires intelligent automation to maximize.
While CPU Busy will always be interesting, the smart move is to focus on the dollars. When you consider that your salary and your job are dependent on how much money the business can make from each transaction, relative to how much that transaction costs, you’ll see the value of putting cost at the top of your metrics “hit list” and ThruPut Manager at the top of your wish list.
Latest posts by Denise Kalm (see all)
- Where You Should Be Driving DevOps Automation on the Mainframe - December 3, 2018
- The DevOps Balancing Act of Measurement and Reward - October 9, 2018
- The Evolving Responsibility of Mainframe Application Testing in DevOps - September 6, 2018