Having direct control over your own storage assets is a responsibility some organisations view as a regrettable burden. They are attracted to the idea of paying for on-premises storage like they would pay for public cloud storage — as if they could give up their personally-owned vehicle for a series of Uber rides.
In a way, Uberised Storage-as-a-Service (STaaS) dumbs down storage purchases. This is the public-cloud model of buying storage as a service and not as a physical array or server-based system with controlling software. Once you subscribe to STaaS, you lose sight of the underlying storage array hardware system and software attributes.
When you buy an Amazon Web Services (AWS) storage facility you choose from a list of service options, not from a list of storage hardware and software types. Thus, if you are looking for block protocol access, “You can choose from six different volume types to balance optimal price and performance. You can achieve single-digit-millisecond latency for high-performance database workloads such as SAP HANA or gigabyte-per-second throughput for large, sequential workloads such as Hadoop. You can change volume types, tune performance, or increase volume size without disrupting your critical applications, so you have cost-effective storage when you need it.”
The volume types include, for example, gp3 general purpose SSD volumes and io2 Block Express. Of course there are physical storage systems underlying these services, but AWS rents out levels of storage service that it delivers from this underlay. It composes the underlying hardware and software, so to speak, to deliver the services and service levels it has sold to its customers.
Is storage a private vehicle or an Uber?
When we decide to use a motorised vehicle to get from A to B we can use our own vehicle — one we have chosen — in which case we pay attention to the engine type, size, cost, power, fuel economy and other aspects of the vehicle’s hardware. Or we use a taxi, in which case we don’t. All we care about is that it has seats, meets legal requirements, picks us up at A and delivers us to B in a reasonable time at a reasonable cost. We give up the choice of whether it is a Porsche Macan or Ford Mustang. This is the Uberisation of personal transport.
When on-premises storage system suppliers start selling storage in the same way as AWS, Azure and GCP — by offering storage types and levels to customers and not the underlying hardware and software — then the underlying platform ceases to matter to the end user.
Startup Zadara, with its virtual private storage array, has always provided storage arrays as a service. But now mainstream incumbent vendors are getting in on the act.
HPE is introducing a public cloud-like operating model. Its DSCC (Data Services Cloud Console) is a STaaS platform with data service software abstraction layers operating on Alletra storage arrays. The DSCC provides access to a Cloud Data Services suite of software subscription services which store, protect and move block data.
Customers buy (subscribe to) storage services, not to Alletra array models with specific attributes.
Dell’s APEX project also involves STaaS with managed file and block storage services due to be provided on-premises this year.
When on-premises storage system suppliers start selling storage in the same way as AWS, Azure and GCP, the number of customers interested in storage platform specifics will decline.
The vendor will ship hardware and software to you, but you won’t be concerned about the array model number or the scale-out storage server CPU core count and DRAM capacity — because that’s not the level at which you are buying, at which you are making your choices.
As long as the vendor-shipped kit meets the service level agreements in terms of access protocol, capacity, IOPS, bandwidth, latency, data resiliency, etc., you are happy. Or you should be, because that’s the level at which you buy — not physical attributes but service attributes.
We say vendor-shipped kit, but STaaS can be supplied as a public cloud service from a vendor’s own public cloud.
A logical development would be a hybrid STaaS combining the on-premises and public cloud worlds. You could, in a theoretical example, buy a set of Cloud Data Services from HPE which combine on-premises and public cloud elements, with HPE contracted to provide the services — be they on-premises or using a public cloud supplier’s platform.
In theory, Pure Storage could do this, combining its on-premises FlashArray platform and its Cloud Block Store public cloud offering.
One of the features of STaaS, like Uber, is short-term cost control and long-term cost unpredictability. You know how much an individual Uber journey will cost, but not how much you will spend on Uber over a quarter. That’s because of surge pricing and not knowing how many journeys or of what length you will take.
Public cloud STaaS also has unpredictable costs over multi-month periods, due to add-on costs such as egress fees, usage-based costs, threshold-crossing price increments and protection safeguards across zones and regions. Public cloud sticker shock is a known thing, and cloud costs can mount so high that repatriation becomes an option — witness the Andreessen Horowitz Trillion Dollar Paradox article.
Egress fees can bear little relation to reality, with fantastic price markups, as CloudFlare documented. Lock-in can bring tremendous disadvantages.
You do know how much physical on-premises storage hardware and software costs, because you bought it — lock, stock and barrel.
Will on-premises STaaS become popular? In a way, it is a form of outsourcing and, like that scheme, will appeal more to businesses needing to save cost and conserve cash. They move from a CAPEX model to a pay-as-you-go OPEX model. Once a STaaS customer gives up buying their own on-premises storage kit, though, they lose the ability to control their own storage destiny. They can only go where their STaaS supplier lets them go.