Home > Blogs > Virtual Reality


Identifying Questionable Assumptions in the Amazon TCO Comparison Calculator

Amazon recently launched a new version of their Total Cost of Ownerships (TCO) Calculator that compares VMware on-premises solutions to Amazon Web Services (AWS) offerings. Our many customers choose us as their infrastructure platform and stay with us because we provide the best value. The Amazon calculator tries to create a different perception by using biased and inaccurate assumptions.

Stacking the Deck…Obviously

Amazon claims their calculator provides an “apples-to-apples” comparison, but in reality, it doesn’t come close to doing so. Their calculator contains biased assumptions regarding VMware’s TCO, which inflate the costs of an on-premises cloud and underestimate the true costs of using a public cloud solution.

For instance, Amazon’s calculator:

  1. Assumes a customer has no existing on-premises investment. Cost assumptions include the purchase of entirely new data center space, racks, networking switches, spare parts, etc., which would not necessarily apply to a customer making an incremental investment in their IT infrastructure.
  2. Compares VMware vSphere Enterprise Plus, our most feature-rich edition, against AWS infrastructure. vSphere Enterprise Plus includes features such as vMotion and High Availability that Amazon customers do not get. As a result, AWS customers often find they have to re-architect their applications in order to work around these missing capabilities. Beyond the basic hypervisor features, Enterprise Plus also offers Application High Availability, Storage vMotion, DRS and Storage DRS, a fully manageable distributed switch, and storage and network I/O control. Customers can get some of these features in AWS by adding on 3rd party solutions, but features like I/O control and DRS are not features that can be added on AWS. Without these capabilities, AWS lacks an effective noisy neighbor solution, forcing customers to seek other ways to manage their applications and performance. These missing features are all part of the hidden costs customers encounter when they switch to AWS and one of the key reasons they delay or cancel AWS migration projects.
  3. Assumes high hardware costs. The calculator assumes a cost of $7,851 for a server with 2, 6-core CPUs and 96GB of RAM (including support). For instance, the same configuration at Dell can be priced at $4651. These price gaps add up, especially when considering larger environments.
  4. Does not include AWS support costs. The calculator includes the cost of VMware’s highly regarded Support and Subscription service for one year whereas no costs for AWS support are included. AWS support fees can be sizable.
  5. Most IT shops are keeping their hardware longer than three years. At the end of the three-year time horizon, the VMware user owns their infrastructure and VMware software licenses. The AWS TCO Calculator truncates the comparison at the three-year mark, yet operating VMware on your on-premises data center can lower your TCO over the long-term.  It is inaccurate of Amazon’s to assume that the value of the entire private cloud investment vanishes after three years. (We would love to hear about your real-world experiences, with your hardware’s working lifespans, in the comments section.)

Another Take on VMware vs. AWS TCO: VMware’s Own TCO Calculations

We decided to take a look at how costs might look using our math. The following is a VMware version of the TCO comparison against AWS. It compares costs associated with running conventional workloads on AWS and VMware infrastructure.

Conventional Workloads TCO Comparison

In a separate VMware TCO comparison calculation for a 100 VM environment, VMware TCO is $394K compared to AWS’s $487K over a four-year period. This represents a 21% cost savings when choosing VMware.

AWS TCO Comparison

This comparison uses the following 100 VMs for AWS:

  • 50 “Small” Windows servers (1 vCPU, 4 GB RAM, only full utilized about 4 hours per day; “Light Reserved” instance)
  • 25 “Large” Windows servers (4 vCPUs, 32 GB RAM, fully utilized about 18 hours per day; “Heavy Reserved” instance)
  • 20 “Small” MS SQL servers (1 vCPU, 2 GB RAM, only fully utilized about 4 hrs per day, “Light Reserved” instance)
  • 5 “High MEM Extra Large” MS SQL Servers (2 vCPU, 16 GB RAM, 400 provisioned [guaranteed] IOPS per VM, utilized about 12 hrs. per day; “Heavy Reserved” instance)
  • For the AWS deployment, it also included Business support (24×7 phone support) and the VPC service, which provides a security perimeter for the VMs.
  • For the VMware cost it uses the ROI/TCO tool (http://roitco.vmware.com) to estimate the cost of running 100 VMs.
  • It uses vSOM Standard for the infrastructure running on 5 ESXi hosts with 2 CPUs, 6 cores per CPU, 128 GB RAM per host and 6 NICs per host. Also included in this deployment is 10 TB of SAN storage using iSCSI.

Note that for this sample environment, the calculations assumed licenses for vSphere with Operations Management (vSOM) Standard, which offer more features and functionality than that of AWS and contain the features a customer truly needs for this scale environment. There are also additional AWS fees for things such as: data transfer, IP addresses, service monitoring, CloudWatch, etc. which are not captured in this TCO, but are a necessary part of running an application on AWS.

Conclusion

Clearly the AWS TCO Calculator does not represent a fair, “apples-to-apples” portrayal of the costs of an on-premises solution. Amazon’s calculator is underestimating AWS costs and overstating VMware costs. The costs of AWS instances are not the only factor to consider when choosing where to host workloads. Designing for AWS requires developer teams to significantly redesign their applications to account for the limitations and the quality of AWS infrastructure. With VMware, you have access to cost-effective, highly automated, secure infrastructure with a level of control and quality that provides superior value to IT and business units.

With the addition of vCloud Hybrid Service (vCHS), VMware now offers customers a public cloud option with faster time to value and the ability to add or reduce capacity dynamically through the use of hybrid, off-premises data centers. The combination of on-premises vSphere or vCloud Suite infrastructure with cloud-based infrastructure hosted on vCloud Hybrid Services or a vCloud Powered partner clearly provides the best hybrid cloud experience. With infrastructure running on a common technology platform (vSphere) and integrations with existing tools like vCenter, vCenter Operations, and vCloud Automation Center, VMware customers get all the benefits of a true hybrid cloud.

Edit: An earlier version of this post claimed that the VMware TCO was over a three-year time period. The correct time horizon of the VMware TCO is four years. The post has been updated to reflect this change. 

 

 

 

9 thoughts on “Identifying Questionable Assumptions in the Amazon TCO Comparison Calculator

  1. Daniel Lemire

    Great points are made here, however regarding the 3 year time horizon on server hardware:

    3 years I find to be an appropriate value. 4 years is actually quite painful due to improved processing capability and features. You can do 4 years but I believe anything beyond 4 years to be a bad idea and should considered obsolete. At that age it begins to limit the virtualization management feature set due to processor feature set .

    I would like to know what the hardware horizon is for AWS.

  2. Craig Hosang Post author

    Thanks for your comments. AWS calculated their time horizon for the TCO as 3 years. In terms of AWS’s actual retirement of physical hardware, that’s a little unclear. You can still utilize older generation AWS instances (http://aws.amazon.com/ec2/previous-generation/); however hardware refresh rates are not publicly documented.

  3. Pingback: One certainty in cloud: the fight is on — Tech News and Analysis

  4. Pingback: One certainty in all this cloud confusion: the fight is on | Essential Post

  5. Pingback: One certainty in all this cloud confusion: the fight is on | Content Loop

  6. Pingback: One certainty in all this cloud confusion: the fight is on | Tech Auntie

  7. Pingback: Global Tech Review | One certainty in all this cloud confusion: the fight is on

  8. Pingback: One certainty in all this cloud confusion: the fight is on | Earthgrid

  9. Ben Meadowcroft

    One other aspect I noticed was that the TCO tool performs some odd VM to EC2 instance matching at the time of writing. What do I mean by that? Well, assume a VM with 4 vCPU and 16G RAM, using the AWS TCO calculator you have the choice to optimize by memory and CPU (but not both). However this doesn’t necessarily lead to the best match for your VM, for example:

    - When optimizing by memory AWS TCO matches this to:- c3.xlarge with 4 vCPU but just 7G of RAM
    - When optimizing by vCPU AWS TCO matches this to:- r3.large with just 2 vCPU and 15G of RAM

    The calculator seems to match on either RAM or vCPU but not both. A better match for my VMware workload would have been an m3.xlarge but that is more expensive than either of the options they selected.
    c3.xlarge = $0.210 per Hour
    r3.large = $0.175 per Hour
    m3.xlarge = $0.280 per Hour (best match)

    If you aren’t actually matching like with like then it’s easy to claim additional savings (in this case an extra 30-60% on the compute piece) that aren’t there in reality.

Comments are closed.