Insights

A Reserved Instance (RI) is a type of pricing option for certain cloud computing services, such as Amazon Web Services (AWS) and Microsoft Azure. When you buy a reserved instance, you are committing to using a specific type of compute instance or other compute resource for a specified period, usually one or three years. In return, you receive a discount on the hourly usage rate for that resource. 

Reserved instances can be a good choice if you have a predictable workload that requires a certain type of compute resource for a long period of time. By committing to a reserved instance, you can save money compared to paying the standard, on-demand hourly rate for that resource.

Purchasing RIs is very easy.  It involves a couple of mouse clicks and voila, you have just reduced your overall cloud bill by 20%.  This can lead to an overreliance on RIs to control cloud costs.  It is not unusual to see many organisations with 90% + coverage rates for their compute resources, which is high. This may work for organisations who have aggressive growth or who ensure they right-size their compute resources before they commit to these resources for one or three years. However, this is rarely the case.  

Many organisations migrate to the cloud/build on the cloud at speed, leaving things like right sizing to ‘do another time’ but then never get around to doing them. Time constraints mean these unoptimized compute resources get covered by an RI, leading to a perceived 20% saving from the cloud bill, when in reality, the savings should have been much higher. 

Worse yet, applying this strategy to an environment which is ever evolving could lead to organisations double paying for their compute resources, which I refer to as RI misuse.  When RIs are misused, they prevent you from benefitting from the flexibility of the cloud. You are unable to take advantage of only paying for instances when there is demand and reducing compute resources during periods of low demand or responding to optimisations within your application, whether that be a reduction in the number of instances or a change in compute family type altogether.  It may even restrict your ability to take advantage of newer generations of computing resources, which are usually faster as well as cheaper.  

The more appropriate way to use RIs would be to understand what your minimum compute requirements are and buy enough RIs to cover that requirement. The rest of that demand can be covered by compute savings plans and pay-as-you-go instances.

Flexibility in the Cloud – Not in Your RI Purchases. 

One of the important things about the cloud is the flexibility you have. You only have to pay for what you use when you need it. This means, if you have a peak period, where your demand is higher for a couple of weeks, you will only have to pay for that additional capacity for those couple of weeks.  

If you have a detailed understanding of your demand patterns, you may observe, you only require that additional capacity for a few hours each day, during the peak period, thus drastically reducing the amount of compute capacity you need to pay for. However, these opportunities become lost when you commit to a high percentage of RIs.  

Another way RIs reduce your flexibility, is when you optimise your application. A once CPU bound application may no longer be CPU bound due to the optimisation of your application or release of new functionality, which causes your user base to use your application differently.   A different instance type could ensure more optimal use of compute resources. However, RIs can prevent you from responding to changes to your application immediately, meaning you miss out on keeping your costs at an optimal level. 

This is set to become more of a problem as Microsoft announced they will reduce the level of flexibility associated with their RI purchases. Previously, Microsoft allowed customers who had an Enterprise Agreement, Microsoft Customer Agreement or Microsoft Partner Agreement, to exchange their RIs for a different type.  They even allowed up to $50k of refunds in a rolling 12-month period.  But from 1st January 2024, all these flexibility options will cease to exist, which means once you have made your commitment, there is no reprieve.

Should Your Organisation Commit to a Savings Plan? 

The likes of Azure and AWS are now pushing their customers to make use of Compute Saving Plans as an alternative to RIs. These allow greater levels of flexibility as your compute resources are not tied to a specific location or instance type. The way they work, is you inform the cloud provider how many compute hours you would like to commit to, and they apply an agreed discount to any compute resources you consume.  This means that you no longer need to work at a granular resource level, you can now work at a global level to work out your requirements.  This is great for organisations who have significant compute growth on the cloud because you can commit to an elevated level of coverage, with the safety of knowing you will not overcommit because of your growth.   

This becomes a lot trickier for organisations who have variable demand as the commitment is calculated at an hourly level.   If you committed to 100 compute hours and only used 80 compute hours, those 20 compute hours will be lost, meaning you would still have to pay for those 100 compute hours, even if you didn’t use them. But you wouldn’t occur any additional costs. Equally, if you exceed your 100 compute hours, any added hours consumed will be billed at the normal Pay as you Go rates.

Source: https://azure.microsoft.com/en-gb/pricing/offers/savings-plan-compute/#how-it-works

Again, the temptation could be to go for maximum coverage, but you run the risk of paying for compute hours you don’t need. 

Useful recommendations for Cloud Costs Savings 

• Consider historical compute usage. How has it been trending over the last 12 months?  Consider getting a view from product teams and the business for plans over the next 12 months to help assess potential compute consumption. Factor in right-sizing opportunities and understand if you have opportunities to reduce your compute consumptions outside of business hours prior to committing to RIs or Compute Saving Plans 
• Having the right mix of RIs, compute saving plans, and Pay-as-you-go compute will help keep your cloud costs at an optimal level without sacrificing flexibility.