Cloud Adoption Strategy: Public Cloud or Private Cloud?

Many companies only consider the public cloud model when considering a digital (cloud) transformation. In this episode, Kumulus Technologies CTO and Founder, Robert Starmer, argues that there are cases when a private cloud will provide greater value to your business. Robert takes a closer look at some of the assumptions that have helped public cloud gain a ‘default status’ in among those that are hoping to save money by moving to a cloud system. Not all of the value propositions are true though. Many cloud servers are actually more expensive than physical servers, as an example. Click on the video above to hear the full story and be in a better position when evaluating your public-private cloud strategy.

Show Notes
Edited Transcript
For those of you that prefer reading, we’ve run the transcript past our editors for a smoother, cleaner read, below.

There’s been a lot in the news these days about using exclusively public cloud environments for all of your IT infrastructure. Basically moving everything from private data centers or closets… actually probably should move stuff if it’s in a closet… but moving it from a private data center, colo facility, etc. and just moving all those applications into the cloud. Why are we actually just building any of this stuff ourselves anymore?

I think there’s a lot of benefit to that model. Especially given that there are certainly instances where we have hardware that consumes more power, consumes more cooling resources, etc. to run than it is actually providing in terms of benefit to the company. A lot of little companies that have, you know, the old desktop that’s become their server that’s sitting in the closet… with things of that nature, I think the biggest benefit you get from moving that kind of classic infrastructure to the cloud is just the potential for resilience. If nothing else, you can usually map an application like that or a system like that onto a shared storage infrastructure and provide snapshots and backups against that. It’s not as resilient as some of the modern cloud applications but it at least protects the asset which is your company’s data in a more effective fashion than a server sitting in your closet.

Now there are a lot of reasons why that’s not necessarily the best strategy. It really comes down to understanding your data, your data processing utilization, the storage requirements for your data, and then the computational costs of manipulating and working with that data both at the compute level; the compute and storage interaction level. So what kind of storage you’re talking about and also potentially storage interaction across sites. If you have resources in a different location to where your storage actually resides, there are potential costs and just consuming and accessing that storage that people haven’t necessarily thought about when we think about cloud. Now again this is a part of in my view sort of a cloud strategy – the sort of conversations that I expect to have over the following months– I think it does come back to strategy.

First if your CEO comes to you– and maybe you’re the director of IT or the CIO– and says, “We need to go 100% all in the cloud. I want this NoOps business model. I think this is the answer– I read it in Forbes (or what have you).” Actually, don’t know the article on know that Forbes wrote an article on NoOps, I’m just saying that these are the sorts of conversations that seem to be coming up recently. I think the first thing that should be a should be said as a response to that is, “That might be the right answer but why don’t we first understand what is the benefit that cloud is going to provide to us.” Look at any of the Five Minute of Cloud or Kumulus on Cloud that I’ve done on a cloud strategy that are sort of a good basis for that first question that you have to ask yourself.

But, just in a nutshell, cloud strategy really is about understanding how your business is going to be able to make use of the cloud and it’s not just from “oh we’re gonna save cost or will get better feature acceleration.” It’s understanding of how all the pieces of your business are going to interact with those cloud resources.

Maybe you have somebody that’s doing some big data or analytics workloads. Maybe the only place that you can do them is in the public cloud because you need machine learning type resources that you don’t have and you’re probably not quite ready to acquire resources to actually manage. Maybe you should never acquire resources to manage. Again these are decisions that can be made once we understand what the strategy is.

If the company’s entire business is going to be built exclusively on machine learning, it may be that it’s more cost effective to buy your own servers, host them in a Colo, run your own machine learning TPU laden environment– depending on how much of that resource you use. It turns out that the TPU resources– other than the free ones that you get on sort of initial loan from many of the cloud providers– are actually a more costly solution than a standard compute resource.

Now I can I can actually do some machine learning and AI-like tasks on those platforms that I could just not do on actual hardware. It would just take too long, it wouldn’t be useful information by the time I had a result from my algorithm. But that doesn’t mean that’s the right answer until you understand what is your business strategy. If you’re just dabbling in this– trying to understand is this going to help? can we get some meaningful results out of a machine learning process?– I would absolutely go with a public cloud. That’s one of the biggest benefits of going with that class of resource. Do you have data sets that you can move in the public cloud that are big enough and meaningful enough? Do you already have datasets in your own private environment that you’re thinking about moving into the public cloud?

We can talk again a further depth about storage requirements and storage models and storage costs in the public cloud. They are not insignificant and actually many people look at the compute cost and forget about the long-term storage cost. Compute you can turn off. Storage, once you place it into the cloud, you’re not likely to just go delete it. Now in many cases uploads are free. It’s only if you actually try to pull the data down again that you actually pay for access to or consumption of that data or retrieval of that data out of the cloud environment. But you still have to be aware of the fact that it still takes time and so time is– you know the corollary– time is money– well in this case time may well be money in terms of the time it takes to actually get the data into the cloud to be able to process it. Especially if you then decide that you don’t want to leave it there and you want to delete it. There are costs involved in all of these sorts of things and those costs have to be factored into the strategy and then that drives the business architecture.

So again I didn’t want to get too deep into strategy, although of course, I walked down that valley again, I did want to talk about some of the pros for public versus private cloud. I think you know on top of first understanding the consumption model for those resources. When we talk about public versus private again cute compute is usually the first thing that comes up. If you have a defined strategy for how you’re going to consume cloud resources maybe your workloads are very dynamic or are going to become very dynamic as you move into that cloud environment it might be that public cloud makes the most sense because you can turn those services on there off. Again, we have to then map the second piece of that puzzle which is we need to look at storage and storage interaction what are your storage costs for long-term storage. How much of it can you move into some of the lower cost but less accessible. It takes a longer time to actually get access to those resources sorts of cloud storage right? They talk about online, near-line, and then cold-line or glacier storage (in certain vendor’s speak) but we need to understand what those costs are and how they seem to map back and forth into the overall system cost.

There are also network costs that we want to factor in. How much does an IP address so that a remote customer can get access to your system and load balancing resources and even inter-site resources for site replication and transactions? Do you have two databases that are in different sites for resiliency, which is a good practice, but that also means that you’re now transiting between those different sites or regions. And there’s a cost for every byte that gets sent across that link. Are you aware of those costs and have you factored them into the cost of what this cloud savings is going to look like?

Again this assumes that your strategy is we’re using the cloud for saving money– might be appropriate depending on use case depending, on consumption in the past the analysis. Analyses that I’ve done indicated that if you’re running a hundred and fifty or two hundred four to eight core class machines with 32 GB or more of memory in them, and usually that then maps to probably a couple of hundreds of terabytes of storage associated with those resource classes. In that model, a private cloud– if those resources are effectively fully utilized — always on maybe running a bunch of VMs or now today maybe running a lot more containerized workloads if that’s your model, it might actually be more cost effective to have two data center locations, run the servers in those data center locations have the the shared link between them to provide resiliency, provide the you know get the bandwidth cost for a public internet access into those environments, manage the load balancers as a part of your application environment, and hire the two to four staff members needed to actually operate that environment assuming that you’re following modern operations practices and automating all those tasks not trying to do all the stuff manually. It might be that that’s actually a more cost-effective approach than going to a public cloud even.

So it is important to understand what those costs are at the scale that your business is at or that your business is aspiring to and working towards in terms of helping you to make that decision. You can also think about the inverse model- public cloud to begin and private cloud as you scale to a certain point. There’s nothing that says that that is a wrong model. Even today, even though again, there’s this big push to say, “Oh and let’s just get rid of all these private cloud things. Let’s go all public.” I mean obviously the public cloud vendors want that and it might be the right solution for your business, I’m not saying it’s not, it’s just that really needs to be mapped into your overall cloud strategy. So don’t give up on private cloud if that’s the right model for your business. Don’t give up on thinking about it as a possible solution for how you might run your business and that’s really what I wanted you to take away today.