IT Pros say this a lot about Azure and AWS. It’s true that IaaS is the most expensive way to use cloud services. Conversely, SaaS and PaaS keep costs lower due to economies of scale. Simply lifting and shifting infrastructure and VMs doesn’t change the math much.
Cost Savings Scale
There are ways to get ROI out of IaaS in the public cloud, however. Here’s a few:
Pick your spots
In some cases, IaaS is a bargain compared to other options. Take DR. Often, an IT department pays for a second data center or collocation facility for DR, or they pay a partner to host and maintain backup systems. That yearly cost can be much greater than using DR services from the public cloud, since IaaS does use shared network, HVAC and other resources spread over many customers.
Gartner Group agrees, saying in their 2016 Magic Quadrant for DRaaS that “IaaS can drive signiﬁcant cost savings when customers have short-term, seasonal, disaster recovery or batch-computing needs.”
Azure Site Recovery is on Gartner’s Magic Quadrant, and makes an especially compelling business case. Customers only pay for the data being stored, and a small monthly service fee. Instead of paying for a hosted system at a second data center 24x7x365, the standby system in Azure is activated on demand.
- There are two ways to pay for VMs: As-you-go or with Reserved Instances (prepaid commitments). Once a deployment scales beyond a pilot, Reserved Instances can save 72% over pay-as-you-go.
- It’s often possible to reduce the costs that comprise a VM in Azure. Fees include the OS, network, attributes of the machine (RAM and CPU). Microsoft Enterprise Agreement customers are able to use Azure Hybrid Use Rights to transfer spare on-premises Windows Server OS licenses into Azure, and cut that cost by as much as 40%. Combined with reserved instances, Hybrid Use Rights can provide 82% lower costs than pay-as-you go. SQL licenses that are owned can be transferred as well.
- Most IT pros over-engineered their on-premises servers for peak usage, 24x7. When building VMs in the cloud, it’s often unnecessary, and it’s always expensive, to take the same approach. Microsoft estimates that on-premises servers are underutilized (6% avg CPU utilization, collectively), and that rightsizing the machine can save as much as 70% (i.e. by using a D12 instead of a D14).
- Many times, costs spin out of control as multiple engineers/departments are spinning up servers. Develop a governance plan that includes an approval process, and/or tag servers with identifiers to make cost recovery possible. Information on tags should include the cost center that’s paying, what apps are running, project identifiers, and who is responsible for technical oversight.
- A new B-Series CPU platform allows you to store credits when the system is underutilized, and spend the credits during peak periods, bursting to 10x what you’ve contracted for.
- Finally, since PaaS is so much more cost effective, consider if that SQL database you’re planning to move must be on IaaS with Azure storage, and move it to PaaS if not.
Once a VM is running in Azure, there are opportunities to optimize and automate to minimize sticker shock at billing time.
- Azure Cost Management and Cloudyn are available tools to assist in ongoing cost controls. Ticketmaster used Cloudyn to improve budget variance and inventory forecasting improved by 42-64%.
- Such tools will show that running VMs on-demand is the number one opportunity to save. That may mean auto-shutting down development or other non-production servers on the weekend (up to 38% savings), or on the weeknights too (saving up to 60%).
- Autoscaling allows a single VM to be turned off when its farm is running at low utilization.
- Resizing VMs allows you to assign appropriate resources (down or up) if a machine is underutilized or taxed.
- Deallocate retired VMs from the environment through the portal. Simply shutting down the VM into a stopped state is not enough. If you retain the data storage, those costs will persist.
- Monitoring using the Billing Accounts Portal and then alerting the owners of anomalous over-expenditures can allow them to rectify an issue before the bill arrives.
- Billing by the minute, not the hour, can shave 5-10%.
Even if you take all of the precautions outlined here, it’ll be cost effective to lift and shift only a small percentage of applications. Microsoft’s own transition of thousands of their own applications is shown below. 45% of their machines, in total, went to PaaS and IaaS, whereas retiring or eliminating duplications was possible for a significant 30%. In summary, use SaaS and PaaS when you can, IaaS when you must, and in advantageous cases like Disaster Recovery.