Back in the olden days, when companies had to buy lots of shrink-wrapped software, vendors used bundles of four or five or more applications — Microsoft Office or Back Office, Lotus SmartSuite — as well as volume discounts to sell a boatload of bits at a big discount. People bought, but they also ended up with rights to a lot more software than they needed (or ever used).
A common sight in those days was stockpiles of actual boxes — shrink-wrap unbroken — in storerooms, bought and paid for, but unused. Later, when companies started deploying from servers, the piles may have disappeared, but the impact was the same: A lot of those software licenses lay fallow. This might have been good for the vendor initially, but it was a ticking time bomb. Customers started to wonder just how great a deal they really got and (god forbid) to question the value of the software itself.
One of the supposed advantages of cloud computing over that on-premises deployment model was that you would only buy what you need and pay for what you use. But a funny thing happened en route to cloud: It turns out that customers still buy too much stuff — they “over-provision” or buy more cloud resources than they need and end up with shelfware, only in someone else’s cloud.
Given the low cost of many of these resources, you might say, big deal. But waste is waste and underutilized resources are a no-no from a security perspective, too. Upfront Ventures general partner Mark Suster has an interesting take on the dangers of shelfware here.
Shelfware in the cloud
There are a couple common scenarios that lead to shelfware in the cloud era. First, many companies often book and pay for long-term instances, which require the buyer to commit for a one- or three-year term to get a bigger discount. The problem is those instances often end up underused or even unused for a big chunk of that time, said the CTO of a San Francisco startup who did not want to be quoted.
Second, in big companies many users are authorized to spin up cloud resources. They may use on-demand instances that carry no long-term commitment, but they don’t clean up after themselves when the job is done. They forget to turn off the resources and the meter’s still running.
And third, many businesses still don’t understand that they need to structure their applications and deployments differently for the elasticity of cloud. “They end up launching servers and letting them live forever like you would in a traditional data center, doing updates, managing user accounts, upgrading software, etc.,” the CTO said via email. “Instead they should be leveraging autoscale, measuring their performance and adjusting capacity as needed. [This is a] very common problem among enterprises moving to the cloud where their staff isn’t yet adjusted.”
Wanted: discounts based on usage, not buy-in
Sometime in the last few weeks, Microsoft turned off access to six- and 12- month discounts for Azure to new users without explanation. A long-time Microsoft partner surmised that the vendor’s salespeople are supposed to encourage actual utilization, not revenue from shelfware and deals like this one tend to engender shelfware.
My guess is Microsoft plans to offer something akin to Google’s “sustained use discount,” unveiled in March. That discount kicks in automatically when a compute instance is used for 25 percent of the time over a month (see chart below.) That idea resonated with startups that are sick of juggling spreadsheets and checking dashboards to optimize their cloud spending. Still, given that Microsoft and (I think IBM as well) is including discounted cloud in enterprise licenses that include other software, the prospect for more shelfware sprawl could be growing.
The problem of shelfware is not lost on vendors. A raft of startups — Cloudyn, Cloudability, Newvem (now part of Datapipe), as well as Rightscale — all help customers pinpoint unused cloud capability. Even the cloud providers themselves see overprovisioning as an issue. Amazon Web Services’ Trusted Advisor, for example, also highlights underused resources for cost-optimization purposes.
Cloudyn co-founder Vittaly Tavor said the Amazon “goes to great lengths to say it’s trying to save customers money by pointing out instances suspected to be idle and by pushing customers to purchase Reserved Instances as a way to save money.” It’s not being totally selfless here, he added, because by pushing Reserved Instances, it gets customers to commit to long term use.
So generally, the bottom line is that customers have to be smart about what they deploy, no matter where they deploy it. Public cloud, in-house server room — all the same issue. Tavor said the over-provisioning in cloud is “just a notch lower than in a traditional data center.”
Gartner cloud analyst Lydia Leong agreed. “The real problem tends to be customers over-provisioning, wasting capacity in the same way that they do on-premises. This can lead to wildly inaccurate estimates about what it costs to run an application in a variety of ways — whether on IaaS, in colo, in hosting, on-prem, etc.,” she said via email.
The more things change…