What you need to know before using the AWS TCO Calculator

 In Accelerite Blog

For those who want to compare the cost of running applications in an on-premises or colocation environment to AWS, the AWS TCO Calculator is a good starting point. However, before making your decision about moving your applications to AWS based on cost savings shown by the calculator, you need to understand what the model used in the calculator covers and what it doesn’t.

This calculator is also a good resource to understand what all you’ll need to take care of if you want to run a private cloud, as Amazon has taken great efforts to cover all aspects of an on-premises environment that have a cost associated with them. However, you can’t say the same when it comes to calculating the cost of setting up an environment on AWS. You’ll notice this after reading all the assumptions and minor details mentioned on the report summary page. The requirements for each customer will vary a lot, and hence Amazon has made some assumptions to make the calculator as generic as possible. It would be helpful to understand what those assumptions are and what may or may not be applicable to your organization before considering the summary report in your decision-making process.

Using the calculator, you can indicate how many servers – physical or virtualized- and storage you currently have. The calculator does let you provide additional details related to your environment, but to keep things simple, it does not ask for a lot of information. Based on the information you provide, you’ll get a summary report showing how much you will save over three years by moving your infrastructure to AWS. The amount could be negative (i.e. you may end up paying higher in AWS) in some cases – more on this later.

Let’s dig deeper into the AWS pricing model, starting with on-premises cost calculations.

  1. Server Costs:
    • The calculator assumes that your servers have 8 core CPUs with 96 GB/256 GB RAM. So, you may end up requiring 50 such physical server machines, when you can get the same compute power by using 25 physical servers with 16 core CPUs and 256 GB/512 GB RAM at the same cost. Reducing the server count will reduce rack space, associated overheads and in-turn the on-premises cost.
    • Rack space utilization is assumed to be 75% of capacity i.e. only 32U are used out of 42U. So,using high compute density servers will help you save on rack space. Also, your rack space utilization pattern may be different for you and will change your overall cost.
    • Hot spare capacity (assumed to be 5% of the total server count per year) generally provides reliability and high-availability in data center environments. Spare capacity is calculated for 3 years (a total of 15%). So, make sure that while providing information related to on-premises servers, you aren’t adding this spare capacity count as that would duplicate the cost. Also, you may have factored in the DR setup count in the input and the calculator doesn’t have any way for you to specify that.
    • Overheads are assumed to be 33%-35% of the server hardware cost and include DC floor space, electricity cost with redundant power supply, etc. However, references considered in the calculator are US based only (changing the currency or the region does not change the reference). The cost will vary based on the country you are located in (and is likely to be much lower in many cases).
    • Admin / IT labor cost is assumed to be a part of overhead costs (most likely with US Salary), but its exact percentage is not shown explicitly in the calculations (the reference page for salaries-http://datacenterpeople.org/wp-content/uploads/2013/03/Salary-survey-FINAL.pdf,  currently throws a ‘not found’ error). Average network admin efforts are assumed to be 8% of total IT administration effort based on the reference report from 2013.
  2. Storage cost:
    • The storage capacity you specify is considered as ‘raw capacity’. This ‘raw capacity’ is initially reduced by 7% on account of OS overheads and further reduced by 50% assuming a RAID-10 configuration. So in the model, the ‘usable capacity’ you are left with is less than 50% of the raw capacity you specified.
    • For calculating backup costs, backup is assumed to be ‘raw capacity’ and not ‘usable capacity’. The calculator doesn’t check if your specifications already include backup capacity hence you need to ensure that you provide accurate data.
    • Prices for SAN and Object Storage are calculated per GB, whereas if you calculate the same on a per TB basis because your storage requirements are in several TBs or more, the prices will drop significantly. However, the calculator does mention that the model assumes higher prices to account for the price of Host Bus Adapters (HBA), Fiber Channel Adapters, Optical or Fiber Channel Cables and other storage equipment.
    • The reference link for storage prices is from 2013. Storage prices have dropped significantly from 2013 till date (and the same trend can be seen from 2011 to 2013 in the reference link itself). So today, you can get storage at a much cheaper rate than reflected by the calculator.
    • Storage cost includes an overhead that is 29% of the storage hardware cost and covers DC floor space, electricity costs with redundant power supply, etc. Refer to related points in server costs above.
  3. Network Cost:
    • Network hardware and software costs are approximated to 20% of server and rack hardware cost. An additional 15% (of network hardware cost) per year is calculated for annual maintenance (so 45% over 3 years). However, your specific hardware cost may vary based on the components you choose.
    • The calculator assumes that 20% of network traffic goes in and out of the DC. Though this is a good approximation, your actual traffic may vary based on applications hosted in your DC, and hence the bandwidth required and the price you pay for the same will also vary.

When it comes to AWS cost calculations, a similar level of detail is not provided for every component that is needed to run your application environment in AWS. Let’s look at them one by one.

  1. EC2 Cost:
    • The calculator assumes a 3-year upfront payment, which means that you are going to reserve all the VM instances for 3 years. If you choose not to reserve it for 3 years and go with a 1-year reservation or on-demand only, then your cost is going to rise significantly. Typically, an on-demand instance is at least 2.5 times higher as compared to a 3-year reservation, while 1-year reservation is 1.5 times higher than the 3-year plan.
    • The input you provide to the calculator is in terms of core CPU, whereas the VM you get on AWS uses vCPU. In AWS terms, each vCPU is a hyperthread of an Intel Xeon core, which has an HT factor of 2. This means an AWS VM with 2 CPUs is internally using a single core on a physical CPU.
    • If you have DR requirements and want to set them up in AWS, you need to choose different regions for the same. While resource (compute/storage) costs may be close to the costs shown by the calculator (since the resource count remains the same, it is just divided across regions), using multiple regions will increase your data transfer costs, which is not taken into account in the model used by the AWS Calculator.
  2. Storage:
    • The calculator assumes storage capacity to be equal to the ‘usable’ capacity calculated for an on premises setup based on the input provided (and not equal to the actual value of the input). If you want the capacity to be equal to the actual input provided, your storage cost will be doubled.
    • Backup requirements are not taken into consideration in AWS. If you need backup in AWS, you’ll need to spend an extra amount on top of what’s already shown by the calculator.
    • There’s a cost associated with snapshots and only one snapshot is taken into consideration over a 3- year duration in AWS costing.
  3. Networking:
    • The calculator takes into account costs for only compute, storage, and business level support. Networking, which is given separate attention in on-premises cost calculations, is missing from the AWS cost. You will likely need
      1. Firewall
      2. Load Balancer(s), depending on load on app(s)
      3. Elastic IP(s) for public facing app(s), if any

      Using these resources in AWS will increase your costs, which doesn’t show up in the model.

    • You’ll need Internet connectivity to access your environment in AWS too. The bandwidth requirement would vary depending on the size of your environment and type of your applications in AWS, and will add up to the total cost.
    • While Amazon doesn’t charge you for data going into AWS, you’ll have to pay for data coming out of AWS. This cost is also not shown in AWS calculator.
  4. Labor Cost:
    • Running applications in AWS will not eliminate all the labor costs. You’ll still need team which will manage and maintain the VM setup in AWS and ensure it’s available whenever required (sometimes 24×7).

In conclusion, while the AWS TCO Calculator gives you good information on various components, the model and assumptions it uses for deriving costs may not match with what you’ll end up paying when using AWS or running your own DC. Additional work is needed to arrive at an accurate cost and hopefully, the points outlined above will be helpful in bringing your estimates closer to reality.

Recommended Posts
Showing 2 comments
  • Dattatray Kulkarni
    Reply

    Good and informative Article, Dattatraya Deshpande.

    I have few comments –

    Elastic IP related costs –
    If an Elastic IP is associated with an EC2 instance and that instance is running, there are no charges for using the Elastic IP.

    Firewall ( Security Group)
    There are no charges associated for using security groups.

    • Dattatrya Deshpande
      Reply

      Hi Datta, thank your for your comment.

      What you have mentioned wrt Elastic IPs and Firewall (Security Groups) is right, but these are very specific usecases. If that’s what your Org needs, then you don’t have to worry about additional cost. But in other scenarios, like organizations wanting to have pool of reserved IPs (non-attached IPs), they would have to pay extra. Also, Firewalls aren’t just about setting security groups, but lot more than that and these additional functionalities come as separate service with extra cost, like AWS Web Application Firewall. AWS-WAF pricing is based on the number of web access control lists (web ACLs) that you create, the number of rules that you add per web ACL, and the number of web requests that you receive. So your cost will go up based on the complexity of rules and traffic volume, but you don’t get any idea of this additional cost just from the model used in TCO Calculator.

Leave a Comment

Start typing and press Enter to search