On-premises infrastructure – as a service

On-premises infrastructure – as a service

Chris EvansCloud, Enterprise, Opinion

At HPE Discover this year, CEO Antonio Neri committed the company to offer the entire HPE product portfolio delivered as services by 2022.  This is a big claim, when you look at the breadth of physical infrastructure HPE currently sells.  At first glance, this seems like a positive move.  However, are the benefits really there and can HPE deliver on this commitment?

Service Delivery

When vendors commit to delivering their products as services, exactly what do they mean?  At its most basic, a service offering means the customer doesn’t own any asset that the vendor uses to deliver the service itself.  This contrasts with capital acquisitions where the customer either buys the technology outright or uses some purchasing scheme to acquire the hardware over time. 

The idea of service-based delivery also applies to software, where we see vendors offering their products on a subscription basis rather than a customer buying a particular version of a piece of software (a standard perpetual licence). 


Initially, there seem to be some good benefits in moving to a service-based model and we’ve seen some of these exemplified by the public cloud.

No capital outlay.  IT departments and businesses don’t need to spend precious capital on infrastructure and risk committing to a sub-standard solution for several years.  Moving to operational expenditure smooths the finances and removes those big budgetary planning and justification discussions.  No-one likes having to justify an unexpected large capital spend.

Staying Current.  Certainly, in software subscriptions, vendors encourage customers to keep on the latest version releases.  This benefits the vendor, with fewer versions to support.  In SaaS-delivered models that don’t sit on-premises, the vendor can upgrade and add new features at any time.  Keeping hardware products up to date in an on-premises delivered solution is harder, however, some vendors have evergreen delivery models that refresh hardware over time. 

Outsourced support.  If the vendor is delivering solutions as a service, then this should also include support and maintenance, delivered on-premises.  Many vendors already do this as part of their product offering, however, typically, standard maintenance is only bundled for a fixed number of years.  After this, rates can increase significantly.  On-going delivery as a service shouldn’t have this year 4/5 sticker shock.


So, moving to a service model makes everything rosy?  Well not really.  There are quite a few downsides to a service-based model, especially for on-premises infrastructure.

You will not own the equipment.  Much like a traditional rental, when you stop paying, the equipment goes back to the vendor.  Consumer IT has quickly moved to the ongoing revenue model as this lock-in can be difficult to break away from.  Think about a subscription to Spotify where all your music “collection” is lost if you stop the monthly payments. Some car purchases on PCP have a similar issue.  They are structured to make it unattractive to make the final balloon payment, incentivising the customer to take out a new contract.  Who wants a 3-year old car anyway, when they can start again and get a new car?

Your overall costs may be higher.  Vendors need to build in additional margin to cover the increased effort in managing a service-based on-premises solution.  This might mean deploying more infrastructure than initially needed, in order to cater for growth.  Vendors don’t want to be visiting customers every week to install new hardware.  Customers don’t want them in the data centre either, so the solution is to build in a buffer – which the customer will be paying for. The decision point here is whether the customer is super efficient at using their hardware (or not). In fact, the vendor may be more efficient, making this point moot.

You can’t sweat the assets.  This issue might not seem a big deal but remember that very few enterprises and medium-sized companies are efficient at technology refresh.  Occasionally, there’s a need to keep equipment on the floor and just pay maintenance, or even dispense with maintenance altogether.  This can be true if IT budgets are stretched, and capital purchasing has to be deferred.  With a service-based contract, it’s not as if you can simply ask the service provider to cut their costs by 20% when you want to save money.

Reducing consumption may not be offered.  Public cloud offers scale-up and down on demand due to the economies of scale they exploit.  However, when physical infrastructure is involved, scaling down isn’t attractive.  A vendor can’t easily remove unused technology from a data centre.  In the case of storage, for example, data might be partially spread across an entire appliance, making it impossible to partially decommission. 

Vendor Deployment Model

How will vendors like HPE deploy infrastructure as a service in a customer data centre?  There are some assumptions to make about the delivery model.

  • There will probably be minimum purchasing capacity and contract terms.  Vendors won’t deploy a single server or 10TB of storage capacity.  They will expect 12 to 36-month terms and minimum quantities (e.g. 100TB of storage).
  • Access requirements may be a challenge.  Vendors may ask for increased physical access to resources, reserve the right to replace resources at their discretion and want the ability to upgrade on demand.  In public cloud these issues would be obfuscated by the cloud service provider, however, with change control and other restrictions, enterprises might have to compromise or accept greater risk.
  • Procurement workflow will need to change.  Customers will need to build in processes to bring new technology into use, with some mechanism to authorise the increases. 

In general, the efficiency of service delivery from the vendor’s perspective will be determined by how flexible their products operate.  Composable infrastructure could prove to be a powerful solution in this aspect, as it would offer the ability to reconfigure and restructure resources more easily.  Modern flash-based storage will certainly be more flexible than HDD or hybrid systems where the platform depends on widespread data dispersion to gain performance. 

Customer Consumption Model

How do things change from the customer’s perspective?  We’ve already highlighted some of the challenges that “as a service” infrastructure can bring.  Probably the biggest obstacle to overcome is the change to internal processes that the model introduces.  Many IT departments haven’t implemented internal service offerings through a standardised service catalogue.  Instead, they work on a project basis, justifying new hardware and leaving central IT with the ongoing maintenance costs. 

Moving to service delivery will mean changing this way of working.  Billing and internal charging need to align to application owners more accurately.  Of course, IT departments could choose not to do this and simply put quotas on consumption, but this seems to defeat the objective of moving to a service-based model in the first place.

One final point to consider is how service levels will be managed.  Will the vendor offer SLAs as part of the service?  If the SLAs aren’t met, what are the penalties?  We talked about the risk of shared service levels in public cloud last year.  Obviously, on-premises deployments will be different but should still meet contractually agreed terms.

Vendor Solutions

Which vendors are delivering using a service model today?  Here are some examples.

HPE offers consumption-based purchasing through their GreenLake offering.  Customers can choose from pre-configured solutions in areas like data protection, databases or Big Data.  Alternatively, GreenLake Flex Capacity allows customers to configure their requirements and receive a quote for the ongoing cost of delivery.  HPE is expected to use the GreenLake model to make good on the 2022 promise.

Pure Storage has been delivering Evergreen Storage for a number of years and was extended in May 2018 into a true on-demand offering.  Customers are charged by space utilised (not provisioned) with minimum terms as short as 12 months.  Ironically, the success of ES2 is one factor being blamed for Pure’s recent earnings drop. 

Dell introduced the concept of the Dell Technologies Cloud at their annual DTW event in 2019.  Customers can order infrastructure based on Dell hardware and VMware software in a solution being called VMware Cloud on Dell EMC.  This is deployed fully managed as “data centre as a service”. 

Hitachi Vantara recently announced both Storage as a Service and Data Protection as a Service offerings for on-premises deployments.  These include in-place upgrades and guaranteed SLAs to meet business requirements. 

Amazon Web Services has announced Outposts, an on-premises fully managed implementation of AWS services such as EC2.  Although neither pricing nor product is available yet, we can imagine this being offered as a service.  AWS has already announced RDS (Relational Database Service) as an on-premises offering.

The Architect’s View

There are some definite advantages in having infrastructure delivered as a service.  For me, the question is one of scale versus the savings that can be made.  Small and medium businesses will see more benefit than enterprises, where the skills and economies of scale kick in.

In some respects, the idea of on-premises service offerings is mirroring the introduction of converged infrastructure solutions some ten years ago.  That market spawned HCI and a range of new product offerings that broke down traditional silos.  As we see on-premises services evolve, I wonder if we’ll see similar developments, as vendors adapt their products to be even more cloud-like than they are today. 

What about HPE? Converting to genuine services by 2022 will be hard, but likely not impossible. I expect we’ll see obvious services like Storage, HCI and Composable being the first to transition. Watch this space and book your slot for HPE Discover 2022 to find out…

Post #7AD9. No reproduction without permission, in part or whole.