InsightaaS: With buzz mounting around Google's entry to Canada - challenging AWS and Azure in hyperscale, and complicating value propositions for local CSPs - we were intrigued by this 451 Research report from Owen Rogers, the analyst behind 451's industry-leading Cloud Price Index.
In the piece below, Rogers compares Google's approach to that used by AWS. In our view (and the Canadian context), this is an apt comparison. Microsoft's expertise in building long-term, mutually-beneficial relationships with its channel partners - and its breadth of presence - means that it is likely to be the platform of choice for resellers and the SMBs that they serve. AWS, by contrast, is touting a model predicated on advanced self-serve technologies that is likely (at least in the near term) to find greatest acceptance within accounts that have limited need for hands-on support: primarily, developers within large enterprises and technology startups/suppliers. Rogers's analysis indicates that Google is also aiming for this community of self-reliant accounts, using approaches that offer improved flexibility. AWS has a lead in both establishing a Canadian presence and in the developer community overall, but Google certainly has the resources and brand visibility to build awareness within this segment.
Google Cloud Next 2017: Is Google's pricing a clear and present danger for AWS?
At its Next summit, the company announced a commitment-based pricing model that rewards consumers with savings of up to 57% with no upfront term payment and flexible, on-the-fly assignment of resources to instances. With an improved free-tier plus an 8% price cut, Google is directly targeting AWS.
The 451 Take
Sustained-use pricing and Google's new commitment-based discounts are easier for users to grasp and execute than AWS' Reserved Instance model, which we think has become bloated with options. However, a pricing model is not long-term differentiation; there is nothing to stop AWS taking a similar approach, or even inventing a new one. Potentially, AWS could automate the rightsizing of virtual machines (VMs) based on workload performance (it already has a Trusted Advisor service which can provide some recommendations), thus making manual selection of a VM size almost redundant and providing a true utility model that balances performance and cost. However, Google still may have an upper hand due to its custom machine types, which makes it possible to rightsize VM CPUs and memory to a tighter fit than is possible with AWS' fixed range of VM sizes. On the other hand, AWS is likely to capture more useful capacity planning data through sales of reserved instances than Google can with its model. As a result, AWS might be in a better position to squeeze every penny out of infrastructure, thereby reducing prices even further. And although Google cites per-minute billing as a differentiator, our mathematical analysis suggests most end users won't save much as a result. So, on balance, our view is that Google's pricing model isn't going to win any enterprises on its own, but it does give the company a tick in its favor when it comes to convenience and ease-of-use. But AWS is likely to respond at some point, either with a new model or a greater discount through better capacity planning.
On-demand pricing enables experimentation; alternative pricing enables cost-effective implementation. The ability to consume resources paying only for what is used is a significant attractor to the cloud, allowing experimentation and evaluation with no commitment and little risk. However, as this experimentation turns into implementation, administrators naturally seek to reduce their costs and establish predictability and availability. At this point alternative pricing methods (particularly for virtual machines) become more attractive, and on-demand pricing becomes relegated to a supporting role. In fact, 34% of the cloud providers in the Cloud Price Index provide best-case pricing.
AWS led the way with Reserved Instances years ago. End users can pay for a full term up front, commit to paying a monthly fee or a combination of both (named All Upfront, No Upfront and Partial Upfront, respectively). In general, percentage savings increase with the length of term (one- or three-year terms are offered) and the size (and therefore on-demand price) of the VM. As expected, consumers that pay up front are rewarded with a bigger discount for that VM, and those that pay on a monthly basis are rewarded with the smallest discount. End users pay for the whole term regardless of how much is consumed, and the average discount for a three-year term is a substantial 60%.
But over the years, reserved instances have become more complicated as the company has tried to better address end-user requirements. Modifications, conversions, term lengths, marketplaces, and regional restrictions have given end-users more options to save money while addressing requirements, but at the expense of simplicity.
Three years ago, Google announced sustained-use pricing where, instead of asking consumers to commit for a discount, Google automatically awards a discount to a virtual machine that shows sustained consumption over the past calendar month. The discounted price is valid for the following month. If the VM is on for 25-50% of a calendar month, the consumer gains 20% off the base price for the following month; the discount is 40% off the base price for 50-75% consumption, and for 75-100% consumption, the discount is 60%. The announcement was a direct response to AWS' reserved instances.
Google's original sustained-use model was incredibly compelling for consumers. It allowed them to benefit from the ability to burst and shrink via on-demand pricing, but with no risk of purchasing resources that are subsequently not used or failing to notice opportunities for cost savings; no up-front payment; plus the freedom to change providers without losing out on prior investments.
At its Next event in San Francisco, Google announced that it was also moving into commitment discounts. Consumers can get Compute Engine VMs at discounts of 57% of standard pricing in return for a one- or three-year commitment, paid monthly. But unlike AWS, end users don't need to commit to a particular size of VM upfront. The user can purchase in advance the number of CPUs and memory units they will require, and can consume these as needed.
This is an easy model to grasp and execute. If a user requires a certain size of virtual machine today, they just size the VM using CPU and memory units they bought in advance. If a user needs a differently sized VM tomorrow, they don't need to change or modify a reserved instance; they simply reassign their prepaid resources to the new VM. And if they need more resources at a point in time, they pay the on-demand price and still benefit from sustained-use discounts. Google's Custom Sizes also allows users to structure the resources in a virtual machine to a granular degree, opposed to having a limited number of fixed sizes from which to choose.
Essentially, Google's commitment model is more fluid than AWS. Google's model acts similarly to a pool of term-resources that can be assigned dynamically to a VM. AWS' reserved instances are somewhat fluid too in that they can be modified in some circumstances, but there are more barriers to reallocating reserved instances to resources due to fixed machine sizes, a modification process and a reserved instance marketplace. AWS' committed resources are less fluid that GCP, and not all users will like this fluidity. Some like paying upfront and knowing how many virtual machines are going to operate in the future. But Google's message is that it rewards loyalty with flexibility.
Shortly after the event, AWS announced Instance Size Flexibility for its reserved instances, which allows some VM's sizes to be exchanged. This increases the liquidity of AWS' reserved instances, but also adds another option and another layer of complexity.
Of course, we don't know how reserved instances or sustained-use pricing is used internally for planning, but we believe reserved instances should provide AWS with a slightly better view of future consumption than Google receives through its pricing model. AWS' model might be less fluid, but the benefit of this approach is that it gives AWS a better defined view of the future which can be used to purchase bulk hardware, negotiate power contracts or even bin-pack resources together. Why should end users care? Because with a better view of the future, AWS may be in a better position to plan for capacity; and with better capacity planning, AWS may have a lower cost-base – savings which can be passed on to consumers. Only time will tell if AWS' approach does provide cost benefits, which could benefit end users in the long run. As you can imagine, a cloud provider's cost-base involves a complex mix of factors, of which capacity planning is just one.
Google has also extended the term of its free tier from 60 days to an AWS-like 12 months, with a $300 credit applicable to any GCP service or API calls. It also rolled out additional Always Free products – non-expiring usage limits that can be used to sample a range of services – including Cloud Engine, Cloud Pub/Sub, Cloud Storage and Cloud Functions. And the company announced a price cut of up to 8% on VMs and continued to moot its per-minute pricing model. We previously proved that unless workloads are likely to live for less than an hour there is little financial benefit in a per-minute model over a per-hour one.
Google is chasing AWS as its primary competition, and is right to do so; AWS is at the forefront of cloud today. Microsoft Azure holds the rear and provides some discounting through the inclusion of Azure in Enterprise Agreements, which provide savings of up to 45% for three-year commitments. Rackspace offers a range of volume, commitment and prepayment discounts up to 33%. Oracle is now taking the IaaS opportunity seriously. Alibaba provides discounts through monthly commitments.
Other providers include: 2nd Watch, Accenture, Adapt, Atlantic.Net, Bit Refinery, BT Global, CenturyLink, CloudSigma, Cosentry, Ctrl4C Cloud Services, Datapipe, Dimension Data, Entel, Exoscale, Expedient, Fujitsu, Gigas, Green Cloud Technologies, Host Europe Group, HOSTING, Huawei, IBM Softlayer, IIJ, Immedion, Internap, Interoute, Joyent, KINX, KIO Networks, LeaseWeb, Linode, Lunacloud, Mandic, Media Temple, NTT Communications, Peak 10, Pulsant, Rackspace, ServerCentral, SINA, Telefónica, TierPoint, UpCloud, Verizon and VMware.
Google's pricing models automatically rewards end users with a discount, are liquid and flexible, and don't require an upfront term commitment. Custom sizes allow an instance to be rightsized to a granular degree and per-minute billing may benefit a minority of users with short-lived workloads.
Google's pricing models are unlikely to give the company detailed visibility into future capacity requirements. Better capacity planning leads to lower costs, which can lead to lower prices for consumers.
It's not rocket science: enterprises like paying less if they can, and a discount that is applied automatically with some flexibility ticks all boxes for cost-conscious CIOs.
There is nothing to stop AWS from copying or even developing its own similar pricing model.
To learn more about the 451 Research Cloud Price Index, please click here