New Year is when people tend to make resolutions for the 365 days to come, like eating more veg, exercising more often, or cutting down on cloud costs.
The latter makes the list as the top initiative for companies for the fifth year in a row. According to the 2021 State of the Cloud Report, over 60% of organizations planned to optimize their cloud costs.
So what goes wrong, and why do so many engineering teams struggle with cloud optimization?
Read on to discover the most common issues that make us waste cloud resources, time, and money, and find out proven solutions to them.
In this guide, you will find:
- Why is cloud cost optimization so challenging?
- The 6 top cloud cost optimization problems to fix in 2022
Why is cloud cost optimization so challenging?
The pay-per-use model of the public cloud has brought more freedom to engineering teams, but this freedom comes at a price. Surprisingly hefty at times, as cloud bill stories from companies like Pinterest demonstrate.
Research proves that most organizations struggle to handle their growing cloud expenses. A typical public cloud spend exceeded budget by an average of 24%.
There are many reasons why it’s still taking place. These often boil down to low cost visibility, complex cloud pricing, and issues with estimating future demand. All these issues may be clouding how much you’re spending and will need to spend (pun intended).
Cloud providers aren’t exactly helping to reduce costs either. Merely deciphering a cloud bill can be daunting. It can be so intimidating that some teams choose to leave their cloud bill to surprise them at the end of the month.
If you don’t want to leave your cloud expenses to fate, read on!
The 6 top cloud cost optimization problems to fix in 2022
#1: Still getting lured by reservations and savings plans
The first thing that comes to mind when thinking of saving on the cloud is to pay less for the services your team uses. Companies choose savings plans or reservations because they come with substantial discounts compared to the on-demand pricing model.
It sounds great to pay upfront for a seemingly predictable cloud spend.
But if you look closely, you’ll see that rather than solving the problem, you’re getting a discount on the issue and committing to it for another few years.
So where’s the catch?
Do you remember how Pinterest had committed to $170 million worth of AWS services in advance but then had to splash out an additional $20 million for extra resources?
Their story helps to illustrate the fact that knowing how much capacity you’ll need in one to three years from now is a tall order.
By committing to one provider for so long, you lose flexibility, get locked in, and may have to pay a hefty price for changing requirements.
How to deal with savings plans?
Well, the best solution is to avoid savings plans altogether!
Don’t purchase resources in advance, but consider approaches that solve the cloud spend like:
- Bin packing
- Resource scheduling.
#2: Falling into the trap of overprovisioning
Overprovisioning takes place when your team chooses resources larger than what you actually need to run your workloads. There is a mindset of safety behind this, as no one wants disruptions in the operation of their apps.
In some business setups, teams are used to getting more resources than their workloads need ‘just in case’. Although performance-wise this approach can make perfect sense to engineers, it takes its toll on cloud waste and costs.
What’s wrong with overprovisioning?
The simplest answer is that it leads to cloud waste and unnecessary expenses that can spiral out of control.
Making overprovisioning a habit in your team is a bad idea in the long run. If you get used to selecting an instance bigger than what your workload needs just to stay on the safe side, think how this plays out as your company and apps grow. You’ll be in for a bill that will cost you an arm and a leg.
Wouldn’t it be better to spend this money on something that matters? For instance, on tackling the climate crisis–a problem to which, by the way, overprovisioning is adding to.
How to deal with overprovisioning?
The usual approach was to invest in custom monitoring and cost management solutions. Their rightsizing recommendations can help reduce your reliance on overprovisioned resources, but you still need to implement them manually.
Another solution is automation. Automated rightsizing can choose the most appropriate instance types and sizes that meet your application’s requirements while keeping costs down.
Whenever a cluster needs extra nodes, the AI-driven instance selection algorithm picks the resources that achieve maximum performance.
And the best thing is that your team doesn’t have to lift a finger as it all happens automatically.
Since overprovisioning is interwoven in the culture of many organizations, it’s essential to build cost optimization routines to balance performance, reliability, and cost. Automation can be a significant step forward on the road to a wider organizational change.
#3: Getting haunted by orphaned cloud resources
It’s easy to spin up an instance for a project and then forget to shut it down.
As a result, many teams struggle with orphaned instances that have no ownership but still continue to generate costs.
Remember Adobe’s mistake, which racked up $80,000 a day in unplanned Azure charges? That’s the type of issue you certainly want to avoid.
This problem is particularly acute in large organizations with numerous initiatives taking place simultaneously, with no centralized resource visibility.
Initiatives managed outside of and without the knowledge of the IT department (shadow IT) can account for 40% of all IT spending at a company. In addition, research shows that shadow cloud usages can be 10x the size of known cloud usage.
What’s wrong with orphaned cloud resources?
Orphaned cloud resources stand for money down the drain with complex sustainability implications.
In short, data centers gobble up a lot of electricity and hardware, contributing significantly to the carbon footprint of the ICT industry. The amount of energy they require doubles every four years, and every new region opening from providers like AWS or Azure contributes to this issue.
That’s why reducing cloud waste is key to halting unnecessary spending and related carbon footprint.
How to deal with this challenge?
Ensuring that you only run the resources you truly need may be challenging, especially in large organizations.
But how do you identify and retire unused instances? That’s where automation comes to the rescue again.
Automated cloud optimization solutions can constantly scan your usage for inefficiencies and compact resources whenever possible. They can also shut down unused instances and processes to reduce your cloud costs.
#4: Managing drops and spikes in demand inefficiently
Engineers building e-commerce infrastructure are well aware of how fast things can change. For example, a single influencer mention can mean millions in new sales or a website that goes down from the surge in traffic.
Most other applications also experience changes in usage over time, but striking a balance between expenses and performance remains an ongoing struggle.
What’s wrong with this?
Traffic spikes can generate a massive and unforeseen cloud bill if you leave your tab open or cause your app to crash if you put strict limits over its resources.
When demand is low, you run the risk of overpaying. And when demand is high, the service you offer to customers may be poor.
Yes, there are cloud cost management solutions monitoring your usage, alerting you in real-time if it exceeds set levels or if there are any anomalies. Such tools can provide you with useful recommendations on adjusting your cloud resources to your current demand.
However, scaling your cloud capacity manually is difficult and time-consuming.
Apart from keeping track of everything that happens in the system, you usually need to take care of:
- Gracefully handling traffic spikes and drops – and scaling resources up and down for each virtual machine across all services you use;
- Ensuring that changes applied to one workload don’t cause any problems in other workloads;
- Configuring and managing resource groups on your own to guarantee that they contain resources suitable for your workloads.
How to fix this issue?
That’s another area where cloud automation can come into play and help you save big on time and money.
Autoscaling can automatically handle all the tasks listed above and keep cloud costs at bay. If you use the container orchestrator Kubernetes, you can benefit from three built-in mechanisms for that.
- Horizontal Pod Autoscaler (HPA) adds or removes pod replicas to match your app’s changing usage. It monitors your application to understand if the number of their replicas should change and calculates if removing or adding them would bring the current value closer to the target.
- Vertical Pod Autoscaler (VPA) increases and reduces CPU and memory resource requests to better align your allocated cluster resources with actual usage.
- Cluster Autoscaler alters the number of nodes in a cluster on supported platforms. If it identifies a node with pods that can be rescheduled to other nodes in the cluster, it evicts them and removes the spare node.
In solutions like CAST AI, you only need to define your horizontal and vertical autoscaling policies, and the autonomous optimization tool will handle the rest for you.
#5: Not tapping into the opportunity of spot instances
Cloud service providers sell their unused capacity at massively reduced prices, especially if you compare the cost with their regular on-demand offer.
In AWS, spot instances are available at up to a 90% discount.
What’s tricky about spot instances?
Since you bid on spare computing resources, you never know how long these capacities will stay available. There are spot instances that come with a predefined duration; for example, AWS offers a type that gives you an uninterrupted time guarantee of up to 6 hours.
But other than that, providers can reclaim the spot instance you’re using, giving you as short of notice as 30 seconds to 2 minutes.
That’s not enough time for a human to react. Creating a new VM also takes more time than that, so you’re looking at a risk of potential downtime.
That’s why if you decide to use spot instances, you need to accept the fact that interruptions are bound to occur. They clearly aren’t the right choice for workloads that are critical or can’t tolerate them.
How to deal with this challenge?
Despite risks, spot instances work great for services that are stateless and can be scaled out, i.e. have more than one replica. Luckily, most services are stateless in modern architectures, as Kubernetes was designed for this type of setup.
Here’s what the process of working with spot instances can look like:
- You need to qualify your workloads and how well it handles interruptions.
- Then you should examine the instances your vendor has on offer and pick the ones that will work best for your needs. A rule of thumb is to choose less popular instances and check their interruption frequency.
- Now it’s time to set the maximum bid strategically to avoid potential interruption if the price goes up.
- You may also want to consider managing spot instances in groups and request multiple types to increase your chances of getting them filled.
You can complete these steps manually, but to make it all work, prepare for a large number of configuration, setup, and maintenance tasks.
That’s where automation comes in handy again. Apart from the process described above, automated solutions can help immediately when spot instances you need are temporarily unavailable.
Solutions like Spot fallback guarantee capacity by temporarily moving impacted workloads onto on-demand nodes and automatically returning them to spot instances when these become accessible. CAST AI’s feature currently supports EKS, Kops, and GKE clusters.
#6: Delaying the adoption of automated cloud optimization
I’ve already referred to automated cloud optimization throughout this text, and this is for a reason.
If you’ve embraced cloud-native technologies, you’re running Kubernetes, maybe even using modern DevOps approaches – automating this part of your infrastructure is definitely a great idea.
Listed as one of Deloitte’s top trends for 2021 and beyond, cloud automation brings tangible results for IT teams, especially in large enterprise environments.
First of all, it reduces the manual efforts you need to put into configuring virtual machines, creating clusters, or choosing the right resources, and more. This change saves time and frees your engineers to focus on more important tasks, innovating, and using your cloud infrastructure to the fullest.
What is more, automation tools allow for more frequent updates, which are key to the idea of continuous deployment. It also reduces the odds of human-made errors, lowers infrastructure costs, improves the security and resilience of your systems, and enhances backup processes.
Finally, automation lets you gain the visibility of resources in use across the company that would otherwise be difficult to control.
In a nutshell, cloud automation is already becoming the new normal in the tech industry.
What’s wrong with delaying cloud automation?
So if cloud automation brings that many indisputable advantages and seems unavoidable, why aren’t all enterprises jumping at it?
Automation can pose numerous challenges, from the resistance to new solutions to fears that they may cost too much to implement and the need to update existing processes.
As in most digital transformation projects, the key to success lies in people and encouraging change on a human level. And when it comes to the workplace, McKinsey’s research proves that fears about being replaced by technology are rife among workers.
However, the advantages automation presents far outweigh the risks. This was evident, for example, in the 2021 State of DevOps Report, where 97% of the surveyed companies agreed that automation improved the quality of their work.
Delaying cloud automation in 2022 equals missing out on benefits like:
- Choosing the most efficient instance types and sizes for your apps;
- Autoscaling cloud resources to handle spikes and drops in demand;
- Eliminating resources that aren’t being used to cut costs;
- Optimizing spot instances by managing potential interruptions;
- Reducing unnecessary expenses in other areas, such as storage, backups, security, configuration changes, and more – all of that in real-time, and at a fraction of the cost you’d have to spend on manual implementation.
How to fix this challenge?
The problem of overcoming human resistance to change and new solutions is as old as the hills.
Reminding your team that automated cloud optimization will liberate them from the burden of repetitive tasks may not suffice.
Showing and experiencing is always better than telling. So why not try out CAST AI for free to discover which areas of your team’s cloud setup it could automate and how much you could save?
May this New Year bring you love, happiness, and lower cloud bills! I hope you’ll embrace the benefits of automated cloud optimization in 2022.
Leave a reply
I guess savings plans are the only way out card for some people that have yet to explore the lands of cloud automations and optimizations. Thats the only way thay can aim to save more on their spendings in their perspectives
Spot instance management is becoming a popular topic for great cost management using cloud providers since more and more automations can handle interruptions and move workloads faster than anyone else could. Insightful!
Pretty extensive list of things to look in to for developers starting in the cloud world or starting to scale in it, amazing job
We’ve been overprovisioning resources for so long and now looking back in to those days, we are a bit embarrassed to even talk about it but I guess many young companies come across these type of things, thanks for educating others and putting a tap on amazons pockets
I think that this article is very useful for those companies that are struggling to optimize their cloud costs but don’t know how.
Laurent just hit the spot when listing the main issue that we are facing right now – aws savings plan. Yes, it saves us some costs but limits us to the extent that its uncomfortable for the business to move on..
I feel like everyone in my surroundings seems to be looking out for a new “thing” , an automation tool that could tacle your mentioned issues and scalability problems saving us some time. I’ll give CAST a go, seems promising
I found these issues and actions for it very usefull so I wanted to say a little thank you for your effort in spreading the knowladge around!
We’ve always found issues in spike management and it resulted in some pretty nasty situations over time which are giving me flashbacks to this day, but thanks to tools like CAST we can now live stress free
My issue is that I get lost in all of the EC2 instance types but I’m a junior in AWS management, so this tool allows me to preview what it recommends for my clusters and I can do all the work to it afterwards for free!