Sizing and Migration Considerations for POWER9 Cloud Workloads
IBM Champion Tom Huntington discusses common cloud challenges.
By Tom Huntington10/01/2019
The cloud has great curb appeal—it’s the simple solution to 101 complex IT problems. “If I could only move my application to the cloud, it would solve all of my issues.” But is it really that easy?
With newer cloud offerings from IBM and Skytap, who provide the platform only and Google’s statement of direction, IBM POWER* in the cloud has been getting a lot of play in 2019. Yet many established MSPs could—and do—argue that they’ve been hosting IBM i and AIX* workloads in the cloud for years. Not only do you get assistance migrating to the cloud, but MSPs help you manage your environment once you’re in their data center.
3 Cloud Challenges
With or without help from MSPs, I see three challenges here:
- Figuring out how big your cloud VM (partition) should be.
- Migrating both your data and your application into the cloud VM—no small feat when you’re talking terabytes of data!
- Keeping your cloud data in sync with your on-prem data until you’re ready to cut over.
Sizing is a key area where Power Systems* users differ from our friends on Windows*. The size of your cloud VM is less of a concern if you’re on Windows because on-prem Windows footprints aren’t scale-up; they scale-out. IBM Power* environments are mostly scale-up, so properly sizing the number of processors and the amount storage is important for IBM Power in the cloud.
And when it comes to transferring your data and applications, you typically cannot save and restore your data into cloud. Cloud providers don’t want to play with physical tapes. They don’t have tape jockeys and prefer all data to be handled electronically.
Getting Your IBM i or AIX Workload to the Cloud
Let’s tackle the sizing hurdle first: You need to know what you’re currently consuming on-prem before you go rent a VM. Planning is critical, and having the right tools or knowing how to read the historical data is where you start.
IBM i has this great flight recorder called IBM Collection Services. It typically collects your performance data every five or 15 minutes by standard default. If analyzed properly, this data will help predict your peaks and valleys in performance consumption. Yes, the beauty of cloud is that you can just spin up more resources on the fly, but a cost is associated with consuming more than you expected in the cloud. The cloud provider might not mind, but your team approving the move to the cloud definitely will when you explain performance is not good unless you add more processors and more disk space.
In the AIX world, you need to use NMON data. Again, I advise collecting the data all of the time and keeping enough data so you can capture peak workload consumption periods. It helps you predict planned consumption and avoid undersizing your costs in the cloud. NMON and Collection Data can be analyzed by several tools in the market.
With your cloud VM rightsized, you’re ready to look at moving the data. Realistically, you have two options: logical replication that uses journaling technology to establish your data in the cloud and/or hardware replication. Either approach can work for IBM i or AIX and you can use both options simultaneously.
Logical replication is more granular, allowing you the option of just choosing the data and application you need to move while giving you the ability to keep the data synchronized with the on-prem data until you cut over to the cloud. And you won’t need to reconfigure any hardware. Hardware replication (i.e., PowerHA*) is another fine option, but it could potentially require configuration of hardware in the cloud to match your on-prem configuration.
Good Things Come to Those Who Plan
The Power platform is known for its reliability and scalability. You need to plan for both, or you will overrun on costs and underdeliver on reliability, which results in a failed project.
Taking the time to look at your on-premises workload before you leap to the cloud prevents awkward—even incriminating—conversations about gross overages on resource consumption in the cloud. Planning for your peaks and valleys helps you understand how you can take advantage of fluctuations in your consumption to fit your predictions, so you can spin down consumption on the fly and save on subscription charges.
Performance planning should provide you with a monthly plan based on past data, but you need the data first. Start the flight recorders for Collections Services and NMON today and use a tool that can help consolidate this data.
The chief benefit of logical replication is its flexibility to work between different hardware and OS levels. It gives you the ability to transfer entire libraries and directories of data to the cloud and keep in sync as you change it in the on-prem version. This migration software can generally be obtained with a subscription-based license for your desired period of time—you just need to ask the vendors.
Yes, moving to the cloud is attractive for organizations that no longer want to be in the business of managing servers. But you still have to plan. And you have to have a good sizing migration strategy.
Tom Huntington is vice president of Technical Services at Help/Systems Inc. More →