Blog

Price comparison between AWS, Azure, and Google

March 28, 2018 | Ken Leoni

Putting together an IaaS price comparison between AWS, Azure, and Google is a confusing and time-consuming process. While they all deliver public cloud services, their market strategies and business models vary.

This means in addition to assessing the cost of the virtual machine resource, IT Departments will need to work closely with C-level managers to determine

Price comparison between AWS, Azure, and Google

  • Is your organization experimenting or is it all-in for cloud?

  • How large a financial commitment is your organization willing to make and when?

  • Is your organization willing to pay “up-front” or must payments be spread out over time?

  • Which is more important - variable pricing that could decrease/increase or predictable fixed pricing?

  • How important is it to be able to receive a partial refund on unused public cloud resources?

 


Where AWS, Azure, and Google converge

  1. Operating System – No matter the public cloud provider, virtual machine pricing will vary depending on the operating system running inside the virtual machine. Pricing can range from free (i.e. CentoS), to paid (i.e. Red Hat Linux or Windows). One would think that you’d pay the same premium for a given OS across public cloud providers, however, surprisingly this is not the case!

    Your mileage will actually vary depending on Operating System and Public Cloud Provider.  Keep in mind, depending on the provider you may also be able to use your own purchased OS licenses.


  2. Server Instance Type – It isn’t nearly as straight forward as one would think. The physical attributes of the virtual machines are different between AWS, Azure, and Compute Engine. The memory, CPU, and disk configuration options are all similar, but different enough to warrant closer scrutiny, making a one-to-one comparison a bit confusing.  For example, Azure virtual machines include a local SSD for temporary storage, if that is important you’ll need to option that out with the other providers.

    Also, keep in mind that new server instance types are periodically added as server and disk technology continue to advance. So, the number and classification of instance types can also change.

  1. Location - Datacenter location is also a factor, especially if your requirements span multiple continents. The number of locations is constantly expanding – Amazon has 54 worldwide availability zones, Microsoft's is growing out to 50 regions, and Google's fast-tracking to 45

    The more significant part of the equation is to recognize that you will pay differently depending on location. Depending on the cloud provider, the premium can vary significantly from location to location.


 

AWS vs. Azure vs. Google  Download Price Matrix


Where AWS, Azure, and Google diverge

Ultimately what gives IT departments fits is how public cloud providers classify and discount their public cloud resources.  While there are some similarities, the differences can be compelling enough for one to either gravitate towards or move away from a particular cloud provider.  In some respects, the pricing is itself can be secondary to the licensing terms.

 

Amazon EC2 Discount Strategy

One could argue that Amazon’s EC2 discounting strategy is both the most flexible, and the most confusing of the major public cloud providers.

  • On-Demand – As the name implies you can spin up a virtual machine “On-Demand”. This category uses pay as you go pricing and is ideal for short term workloads that require dedicated resources.  There is no discount for this category. On-Demand instances let you pay for compute capacity (based on virtual machine size).

    Linux instances are billed in one second increments, with a 1 minute minimum, while Windows instances are billed per hour.

    These instances are a good starting point for first time cloud deployments, where IT organizations are unsure of the direction they want to take and are unwilling or unable to fund up-front costs.


  • Reserved Instances (RI) – Quite simply, you are making an up-front commitment of 1 year or 3 years for compute capacity of a specific Azure EC2 instance type and in exchange receive a discount over the On-Demand price. That EC2 instance, once in use, is a dedicated instance, making it ideal for constant workloads.

    The discount level will vary depending on the length of the commitment, whether the instance is paid 100% upfront, partially upfront, or monthly. Also “Convertible” Reserved Instances allow for moving between different instance families and operating systems (restrictions apply).

    If you over-commit the number of RIs, Amazon won’t take the instance(s) back or give you a credit. Unneeded or unwanted RIs must be sold to a third-party in the Reserved Instance Marketplace, keep in mind there are a number of restrictions and you also have to give Amazon 12% of the RI’s selling price.

    Over-committing RIs can be costly. Selling RIs in the Reserved Instance Marketplace means a bit of work and no guarantees, but at least you can recoup some of the cost of unwanted RIs.


  • Spot Instances –  If you have a workload that can be run without a specific deadline, Spot Instances are ideal. The workload will leverage idle EC2 capacity. Pricing adjusts gradually based on long-term trends in supply and demand of unused EC2 capacity. Amazon claims savings up to 90% over On-Demand.


  • Dedicated Hosts – A physical server with a fully dedicated EC2 instance.  Users supply their own software (and likely expensive) licenses.  Dedicated hosts also can address compliance and regulatory issues. Pricing is the same as On-Demand or Reserved Instance.


Concept of important choices of a businessman

Microsoft Azure Discount Strategy

Microsoft’s relationship with its customers is different from that of the competition because it owns and controls the Windows technology stack. This presents Microsoft (and it customers) with unique pricing and discounting opportunities.

  • Pay as you go –  Ideal for short-term volatile workloads that can’t be interrupted. Azure users pay for the compute capacity (based on the size of the Azure virtual machine).  As is the case with most (not all) public cloud providers, no up-front commitment means it is the most expensive option.

    Azure bills per-second rounded down to the last minute.


  • Reserved VM Instances – As in AWS, there is a 1 or 3-year commitment towards the use of a certain amount of compute capacity.  Once an Azure RI is in use, it is dedicated and won’t be interrupted.

    There is only one payment option for Azure RIs and it is a single up-front payment.  On the positive side, Microsoft makes it easy for Azure users to exchange RIs at any time - a prorated refund based on the unused RI amount can be applied to the purchase of a new RI

    If you over-commit on the number of Reserved Instances you can cancel and Microsoft will refund the RIs -  you'll receive a prorated refund based on the unused amount minus an early termination fee of 12%. 


  • Azure Hybrid Use Benefit (AHUB) – The licensing of the Windows Operating systems is an area where Microsoft can offer a distinctly unique experience to its customers.

    It is important to understand that no matter the public cloud provider, when calculating the cost of compute capacity – you must include the cost of not only the virtual machine, but the cost of the operating system license as well.  A virtual machine running a free Linux distribution is going to cost quite a bit less than one that is running the Windows operating system.

    The AHUB applies to Windows Server licenses covered with Software Assurance.  It allows organizations to extend the use of their on-premises Windows licenses to Azure. Basically, you’re on-premises licenses can be used in Azure, allowing an Azure Windows VM to run at the base cost of an Azure Linux VM.

    The Azure Hybrid Benefit for Standard Edition licenses can only be used once either on-premises or in Azure. Datacenter Edition benefits allow for simultaneous usage both on-premises and in Azure (so you actually get a two-fer here).  Datacenter Edition certainly makes it quite a bit easier and less disruptive to transition from on-premises to Azure.

    The Azure Hybrid Benefit Savings Calculator shows how to best leverage your Windows licenses that are under Software Assurance.

     
  • Microsoft Enterprise Agreement (EA) – All public cloud providers have the ability to provide special discounting to their Enterprise customers. In many respects Microsoft holds an advantage here because they’ve been active in their accounts for decades and have had EAs in place for longer than some of the public cloud providers have been in business. Their close relationship with the Windows customer base allows Microsoft to get quite creative with the EA’s, especially at renewal time.

 

Google Compute Engine Discount Strategy

When assessing whether to work with a cloud provider or not it is not only about the technology, but how easy is it to do business with the cloud provider.  Working with Google is probably the easiest of the three.  Google has done a good job keeping everything simple and straight forward, as well as offering its own unique capabilities.

  • Sustained Use Discount – Automatically provides for discounts when a VM runs for more than 25% of a month. User need not make an up-front commitment. It is the best of both worlds, providing on-demand usage, no commitment, and a discount level up to 30%. Discounts are quick, clean and simple. 

    Google Compute Engine bills per second, after a 1 minute minimum.


  • Committed Use Discount (CUD) – For an even more substantial discount (up to 57%) users can make 1 or 3-year commitments.  Notice we’re talking “discount” and not “instance” CUDs are similar to AWS Reserved Instances and Azure Reserved VM Instances because of the 1 or 3-year commitment, but that’s about it.   

    Where Google’s CUDs diverge from both AWS and Azure -  you’re pre-purchasing portions of CPU and RAM and not VM instance types. You then decide how to distribute the allocation between your VMs.  Google Compute Engine is also unique in that users aren’t saddled with a defined set of VM instance types, they can create custom machine types.

    Unlike Azure which requires payment upfront, Google bills monthly.  However, the big gotcha – to get CUDs, you must purchase commitments, and these cannot be cancelled.

    You’re looking at a no return policy and there is no secondary market!


  • Preemptible VM Instances - If you have a workload that can be run without a specific deadline, Preemptible VMs are ideal. Google claims a cost savings of up to 80%.  In return for the price break – the VM can be preempted by higher priced instances, also the Preemptible VM is short lived and last for up to a maximum of 24 hours.

    Conceptually Preemptible VMs are similar to AWS’ Spot Instances in that both providers are leveraging spare capacity.  There are some key  differences:

    1. The pricing of Preemptible VMs is fixed, while AWS’ Spot Instances are somewhat variable. Organizations that are more concerned about having a predictable cost structure may well gravitate to Compute Engine.

    2. Preemptible VMs are short lived and terminate after a maximum of 24 hours, while AWS' Spot instances have no such restrictions. Workloads that use Preemptible VMs should be distributed and fault tolerant and not require running continuously on a single instance.

 

Summary

So where do we go from here? We can certainly draw some broad conclusions:


AWS
  - offers the most flexible purchasing options, for example: 100% upfront, partial upfront, or monthly payment plans. A multitude of choices mean if there’s a will, there’s way. If you make a mistake and over-commit on the number RIs you can get some of your investment back, with a bit of work.


Azure - is tough to beat, especially if your organization has a long-standing relationship with Microsoft and you already have an Enterprise Agreement in place. Microsoft is aggressively capitalizing on the close relationship with the Windows customer base and garnering steady support for its hybrid cloud environments.  Over-commit on the number of RIs will still cost you, however Microsoft makes it easy to get a partial refund.

Google - makes it all quite simple, Sustained Use Discounts are automatic with no need to make a 1 or 3-year commitment. If you’re looking to save even more, a Committed Use Discount is a definite option, however a mistake is costly – as in no returns!




Although there are substantive differences between AWS, Azure, and Google, there are enough similarities where we were able to build out a pricing matrix and analyze On-Demand and Reserved Instance pricing. Our pricing matrix, including the conclusions we were able to draw from the comparison, is available for download below.

The matrix includes an analysis of Burstable, Standard, Compute, and High Memory machine instance classes.  In addition the matrix examines pricing for multiple vCPU configurations within an instance class.

 

AWS vs. Azure vs. Google  Download Price Matrix

 

 

We value your privacy and will not display or share your email address

Sign Up for the Blog

Heroix will never sell or redistribute your email address.