Cloud computing has transformed the business landscape, offering unprecedented scalability and flexibility. Managing cloud costs is crucial for optimizing operations and financial planning. The challenge lies in navigating the realm of cloud data warehouse pricing models.
Cloud Pricing Trends and Challenges
In this webinar series, industry experts Arijit Bandyopadhyay from Intel®, Doug Henschen from Constellation Research, and Mark Cusack and Heather Brodbeck from Yellowbrick Data Warehouse explore cloud cost optimization and data warehousing.
In Episode 3, our experts explore cloud data warehouse pricing models and their implications for cost control and performance optimization.
Cloud Cost Management Strategies and Best Practices
Our expert panel shares the knowledge and expertise you need to make informed decisions about cloud data warehouse pricing models. They discuss:
- Consumption-based pricing, serverless options, and subscription models.
- The advantages, challenges, and cost implications of each model.
- The potential challenges of cost control and scalability in consumption-based and serverless models.
- The advantages of flexibility and cost predictability in subscription models.
- The importance of cost control in cloud pricing and strategies to achieve it.
- The importance of transparency, cost savings, and vendors passing on the benefits to customers.
- Why you should start small with on-demand consumption, characterize workloads, and then transition to subscription models for better cost predictability.
Discover how to navigate the complexities, make cost-effective choices, and implement strategies to optimize cloud spend. Watch this video to gain a competitive edge in the evolving cloud pricing landscape.
For more best practices on reducing cloud costs, download the Why Data Warehouses Are Ground Zero for Cloud Cost Optimization report.
So we are seeing different vendors out there that have this pay-for-what-you-use model. Doug, could you talk a little bit about that? And any risks related to that and obviously the benefits that are there, also?
Yeah, you hear about consumption-based pricing, you hear about serverless options or autoscaling.
On the serverless side, it’s important to note that it’s an incomplete answer to cost control. Yes, there are oftentimes spiky workloads. Data warehouse workloads where compute demands go up and down you can’t really anticipate. That’s where serverless capacity or autoscaling is a fit.
But whenever it’s easy to add more data, add more users, add new workloads, scaling always seems to go in one direction, and that is up.
Similarly, on the consumption-based pricing, it’s not really about cost control. The storyline is “pay only for what you use,” but in reality, it’s kind of an open invitation to use more, more, more. As more data is added, as more users gain access to analytics, consumption tends to go in one direction, and again that is up.
So what really helps in our view with cost control is a range of subscription models so you have choices and you have flexibility on what best fits your usage patterns, your per-query and capacity-based models are typical.
All the better if there are multiple discounts for time-based and capacity-based commitments, as well as some flexibility to exceed those levels – may be temporarily or by a certain percentage – without facing punitive pricing.
Okay, thanks. Mark, anything to add on that?
Yeah, and we’re certainly seeing this in conversations at Yellowbrick. Which was one of the reasons why we introduced a blended pricing plan of both on-demand and capacity-based subscription pricing that you could combine within the same deployment as well for the reasons Doug mentioned.
I think not only are companies looking to reduce the magnitude of the overall spend, they want some cost predictability here, which is incredibly important for budgetary planning purposes, as well.
I think to key in more on the serverless side of things as well. You really are exposed to a true black box here when you’re looking at serverless. And to the extent, it goes beyond just the unpredictability of being on pure on-demand, highly reactive sort of spend profile.
But also the fact that some of the serverless cloud data warehousing solutions out there today will introduce more resources during the runtime of a single SQL query. That’s kind of non-deterministic.
It becomes very, very difficult to run a couple of times and to truly understand whether you’re going to pay the same cost per query each time you run it.
I think there’s the black box aspect of it, the unpredictability aspect of it. These are real forces acting against any sort of CFO’s department that’s looking to reign in the controls and get some predictability out of that.
When I speak to customers, a typical way that we introduce Yellowbrick is we encourage them to start small. Perhaps they are using on-demand consumption while they’re characterizing their workload.
To Doug’s point, when you see a fixed capacity component of your workload coming out of that, then move that piece to subscription, you’ll get a much more favorable, predictable sort of spend when you do that. Knowing that you can always burst out of that fixed capacity as and when you need it and when your business requires it. But that’s something that we are getting a lot of traction on at Yellowbrick.
Great. Arijit, back to you. Any thoughts on this topic and what you’re experiencing?
Yeah, I think as an example, the Yellowbrick’s a case, the I3ens or I4is are the instance of choice in AWS. So when you are doing this or the customer has a choice to do this, the aspect of the selection of the instance is an important element on Intel because that is also related to performance, IO-optimized to performance, and network-optimized performance or core.
So, the selection of the instances is a vector. The aspect of which platform you’re choosing from Intel is a vector. FinOps tools have become a norm.
About this topic before we get to the next one. Mark, I’ll just look back to you. Just maybe if you could give a little bit of color. I know you talked earlier about how there’s costs or savings that happen at the vendor level that don’t always pass through to the customer.
Can you talk a little bit about what are the costs then for these serverless services and do we see those passing on to the customer when they’re savings? Or what does that look like?
Well, and I think the answer is no, frankly, because quite often what a lot of SaaS vendors in data warehousing are doing is massively overprovisioning on the AWS side or Azure side or whatever, and in the course, those costs are passed back onto the customer at the end of the day as another sort of markup.
Arijit kind of opened the hood a little bit around what we do in the details around Yellowbrick. But to give you a little bit of flavor into how we’ve approached this – when you spin up data warehousing elastically with Yellowbrick in AWS, Azure, or what have you, we curate the Intel instances that we choose here.
And we’ve deliberately chosen the AWS instances that we support to get the best price-performance out of those instances. And that’s completely transparent, we don’t mark up those prices. You’re literally paying the list price for AWS for that instance that’s underpinning our deployment.
Again, we hide the details, you abstract the details away so the user experience is a SaaS-like experience, but you are running it yourself, which sounds like a massive contradiction in terms. And maybe three or four years ago you really couldn’t get away with that.
If you look at the history of running databases and data warehousing in the cloud four or five years ago, you’d be stitching together EC2 instances and having to manage the infrastructure.
With the advent of Kubernetes services like EKS and AKS, that’s managing the elasticity for us in Yellowbrick. And we mask all of these details and provide a simple SQL interface to allow you to create new compute clusters, expand them, contract them, et cetera, et cetera.
Again, I think we’ve latched onto the technical developments, both on the Intel side and in the cloud most recently, to be able to provide kind of a very, very simple, easy-to-consume data warehousing experience without exposing the details, but definitely just ensuring that customers only pay for what they need to as far as infrastructure is concerned.
Okay, great. Thank you.