When I launched Andover Intel last year, my goal was to present an analyst model driven by buyers of technology, not sellers (as is the case typically). That goal has resulted in a number of situations where what I hear (and say in my blogs and elsewhere) differs from what seems to be the common view. In most cases, these situations are associated with technologies that are hyped up—5G, AI, and the cloud are all good examples.
In the case of the cloud, one very interesting thing that’s come out of buyer discussions is the difference between enterprise users of cloud services, and the way that startups and OTT companies use it. That difference came home to me when looking at an InfoWorld article on how to control runaway cloud costs.
Startups, particularly startups involved in social-media or consumer-OTT missions, have long been a major source of cloud revenues. My data, for example, has consistently shown that AWS got more revenue from that source than from enterprises. Further, my contacts in the world of startups and VCs tell me that the cloud is the preferred hosting strategy there, because it avoids a lot of spending on capital equipment and facilities. From a technical perspective, too, the startup world is different from enterprises because its applications are almost all “front-end” technology, where enterprise applications are typically transaction processing and analytics, with the GUI being a layer on top.
These differences mean that cloud decisions for enterprises and for startups are made from very different perspectives. You could say that the startup is compelled to use the cloud; the question is how to make the best of it. For enterprises, there is (and in both my opinion and theirs, always will be) a core data storage and transaction processing component to applications that will be hosted in the data center not the cloud. The question for enterprises is where the cloud boundary will be placed, and how cloud usage will be optimized for what’s outside it.
When I talk with enterprises on controlling cloud costs, it’s pretty clear that they ask three questions that the InfoWorld article doesn’t ask. First, does this particular application/component belong in the cloud at all. Second, what cloud features would implement a cloud-suitable application optimally, and third, what application model or architecture assures the best near- and long-term cost stability. Enterprises, in short, think the big issue is getting an application divided properly between cloud and data center. Startups, having no data center, see the cloud as the be-all-end-all given.
The big benefit of the cloud is its ability to adapt to highly variable workloads, its “elasticity”. For applications that don’t have that sort of variability, enterprises say that data center hosting is invariably less expensive. It’s the need to size for the peak, and pay for that on the average, that makes data center hosting more expensive. Transaction processing loads are typically not all that variable, and in addition they require core company data resources that are protected by security/sovereignty policies. Thus, enterprises say that you have to start with the user-to-application dialog and trace work inward. In order for the cloud to be optimally useful, some edge/user portion of the workflow has to involve pre-transaction activity, meaning browsing in some form. If there is no such activity, it will be difficult to make the cloud pay back.
When considering “browsing” hosting in the cloud, the article recommendation on reserved instances is valuable. In general, cloud services vary from something that approximates real bare-metal servers that are effectively your own, to a kind of best-efforts as-available resource. If you can slide toward the latter, you can make the cloud more attractive, but enterprises say it’s important to understand how any of the service types impact the chances that resources in the cloud might not be available instantly to carry new load. Many enterprise applications can’t tolerate much “blocking” of activity.
Then there’s data. Where there is such “browsing” activity, the question is what exactly is providing the data to be browsed. All cloud providers offer data hosting, so here enterprises say that you need to ask whether there’s a subset of the core data that describes something the user would review prior to creating a transaction. This might be extracted and hosted in the cloud to support the browsing. Of course, it might also be hosted in the data center and obtained by an inquiry dip into the main database. The right approach will depend on how database hosting costs and access charges to reach the data center core databases would compare.
Enterprises say that their ideal cloud usage strategy is to take some of the “front-end browsing” piece of an application and, with minimal effort, rehost it in the cloud. In some cases, though, the optimum use of both cloud and data center is better achieved if you redo some of the data-center piece. One common example enterprises offer is presenting a “quantity flag” to the cloud database, so that if an order poses a risk that there might not be stock on hand to fill it, the transaction would provide a low-stock warning to prospective buyers. Some enterprises also snapshot elements of the core database, including quantity data, on a regular basis and update the cloud database.
Why is there such a difference in how the article looks at cloud cost optimization and what enterprises say? You can read that in the article, which cites multiple cloud users, all of whom are tech companies who don’t represent the great majority of the enterprise verticals. One source is described as a technology startup, another as a software company, a third an Internet proxy service. These don’t represent the enterprise verticals I draw on for broad market views, and in most cases when I reference what companies have told me in my blogs, I’ll specifically exclude the tech verticals because they are atypical.
Enterprises, in the main, aren’t rushing about to get interviews with publications about their cloud use. They’re not setting up booths in tech trade shows. The startups, the tech startups, tend to be doing both, and more. Their views of the cloud may be technologically advanced, and perfectly valid within the frame of reference of how they view cloud and data center. They’re not so perfect in preparing enterprises for the task of cloud cost optimization, and that’s not good for the enterprises, or for the cloud.
