A COST COMPARISON:

AWS Instances Versus Physical Servers

Download the PDF Version Read On

Executive Summary

For many organisations, using AWS instances is viewed as the only logical choice to tend to their infrastructure needs. While AWS is accepted in the hosting industry as a powerhouse, the reality is there are numerous options in the marketplace that are substantially cheaper. Many of these options would suit certain businesses better, as the majority of organisations using AWS’ services are doing so in an inefficient way. This comparative analysis provides insight into this phenomenon – here are our key findings.

  • Our overall findings show AWS on-demand instances are approximately 300% more expensive than using traditional server based infrastructure.
  • Using AWS reserved instances is approximately 250% more expensive than contracting equivalent physical servers for the same length of time.
  • AWS dedicated hosts are drastically more expensive than dedicated servers, costing approximately 450% more on average.
  • Bandwidth (i.e. “internet traffic”) on AWS is incredibly costly so any workloads with high bandwidth demands will be more expensive. Most hosting companies include 10TB with the price of a dedicated server, whereas it would be £666.80 per month on AWS for 10TB of traffic.
  • For users requiring just one or two small virtual servers, there is a significant number of competitors that are cheaper than AWS – Digital Ocean, Linode, and DediServe for example.
  • Prebuilt physical servers and AWS spot instances are closer in cost, with the outcome depending on the availability and pricing of resources at the time required.
  • For storage, the usage costs for Amazon EFS during one sample month have been illustrated in this guide. 1100GB of usage in April would cost £131.79; however, a NAS server with 14TB of useable space would cost £120 from a leading UK server provider – providing nearly 13 times more storage for less money.
  • A dedicated server running MySQL is approximately 3 to 6 times cheaper than an AWS managed relational database running MySQL.
  • AWS is best-suited to use cases where environments require multi-region resilience and redundancy, but minimal resource requirements, as it minimises management overhead. A complex but small hosting platform would be more cost effective on AWS.
  • Any move to AWS will change the nature of supporting the hosting environment. This guide does not explore the detail behind the different skillsets required for supporting and maintaining an AWS environment versus a more traditional hosted platform but identifies the key considerations.
  • AWS has proprietary services that benefit application developers and cut down the need for large amounts of infrastructure. Applications designed “native” to the public cloud make the most of these services and that is when the benefits of public cloud can be realised. However, there are strategic considerations to be taken into account such as vendor lock-in, access to data, DR plans, and whether the public cloud commercial model works for the application in question. These considerations do not fall within the scope of this guide but have been covered by Happi before – for example, in our multi-cloud guide and by various blog posts.
  • This guide does not take into account the wider requirements for re-designing applications for AWS in order to make best use of AWS proprietary services, which incurs development costs but could see a hosting platform reduction.

Pricey Yet Prospering: How Has AWS Done It?

Amazon Web Services (AWS) finished the fourth quarter of 2017 with a 62 percent share in the cloud computing market. While Microsoft Azure has recently eaten into this share slightly, analysts believe the dip to be temporary and predict AWS is set to dominate the market for years to come.

Its success has been attributed to three key factors:

1. FUNCTIONALITY AND PACE OF INNOVATION

The product is constantly growing – for example, a whopping 1,430 services and features were added over the course of 2017. And AWS is adept at keeping abreast of important trends – for example machine-learning and data analytics functionality have been successfully implemented..

2. PARTNER AND CUSTOMER ECOSYSTEM

AWS has struck big partnership deals with prestigious companies such as Salesforce. Meanwhile, businesses such as Kellogg’s, Brooks Brothers, and GE Oil and Gas use AWS to run SAP – another huge system in itself. The AWS Cloud spans 54 Availability Zones within 18 Geographic Regions and one Local Region around the world, with announced plans for 12 more Availability Zones and four more Regions in Bahrain, Hong Kong SAR, Sweden and a second AWS GovCloud Region in the US.

3. EXPERIENCE

AWS began providing public cloud computing services around 10 years ago, much earlier than any of its competitors. Quite simply, being first has given it a huge head start.

AWS has been the market leader for so long now that organisations often unthinkingly just view it as the go-to cloud provider. Indeed, there is a widespread claim that AWS is always a cheaper solution and one which never requires support.

We put together this guide to test that claim, by examining and comparing the prices of different setups with equivalent dedicated servers.

Here we offer a purely commercial evaluation, rather than an examination of differences at a strategic level.

Cloud Repatriation: High Profile Cases

One of the big recurring themes to feature in tech news over recent years is cloud repatriation: the migration of at least one or more workload from public to private cloud. Dropbox has been the biggest name to emerge – originally starting on AWS, the company has since moved more than 600 petabytes of data from Amazon’s cloud to its own data centres. Their reasons were fairly typical, with a company spokesperson stating:

“While cost is always something that we consider, our goal with our network expansion was to improve performance, reliability, flexibility and control for our users — which we have succeeded in doing.”

By regaining control over its infrastructure, Dropbox has saved a substantial amount of money and benefits from being more in control of its own destiny. The cost savings were recently quantified, with the company publicising that a $74.6 million reduction from its operational expenses has been achieved, in the two years since moving.

In a 2017 company conference, Hewlett Packard Enterprise CEO, Meg Whitman described companies hitting ‘the cloud cliff’, where

“Either for reasons of control, security, performance or cost, the [public cloud] platform they went with is no longer the best option.”

Whitman went onto describe the case of Smartsheet, a SaaS collaboration platform that shifted from a public cloud-centred approach as it needed more control over its infrastructure. As with Dropbox, the company found that as it grew, the escalating costs of its cloud-based model had become a sticking point. And this was a particular concern since the company uses a freemium model to entice paying subscribers.

However, while these high-profile organisations have chosen to migrate their services, there are still a number of organisations that are sticking with AWS, seemingly unaware of the cost-saving opportunities that can be made by replacing instances with servers. This guide helps quantify the cost-savings that can be achieved as a result of migration.

Controls

To ensure like-for-like comparisons between AWS instances and physical servers, the region, where applicable, has been set as London and hosting environments running on Linux have been compared.

AWS costs have been made freely available over the Web by Amazon but appear in US dollars. So, in order to compare with the prices of dedicated servers in the UK, AWS prices have been converted to British pounds, via oanda.com. The exchange rate, at the time of publication, was 1USD to 0.74091 GBP.

Resources: Compute Capacity/Server

So that cost comparisons can be made, this section covers non-proprietary AWS resource products, namely various types of compute capacity/server, a range of storage options, and databases.

Compute Capacity/Server

AWS Elastic Computing Cloud or EC2 is a web service that provides secure, resizable compute capacity in the cloud. AWS itself claims that EC2 is inexpensive and that users “pay a very low rate for the compute you actually consume”. In this section, we test that claim.

AWS sells EC2 in four different ways: on-demand, spot instances, reserved instances, and dedicated hosts.

ON-DEMAND

On-demand services require zero commitment and customers pay by the minute or the hour of uptime, depending on their instances. Virtually all rival companies offer a similar type of service.

The table overleaf compares EC2 instances with the costs of dedicated servers from RapidSwitch, an established and leading UK based server provider. 100GB of hard drive storage (Amazon’s Elastic Block Storage – EBS) and 100GB of bandwidth has been factored in, representing light usage.

Physical servers are not relevant for the smallest workloads so n/a has been indicated for these cases. If a user requires just a single, very small server, there are a large number of VPS providers in the market place that compete with AWS – such as Digital Ocean, Linode, and DediServe.

A commonly accepted method of translating a vCPU in an AWS environment to a physical core is that one vCPU represents half a physical core and this technique has been used below to compare EC2 with dedicated servers. Companies such as Pythian, a global IT services company, have published their test results to illustrate how the 1:0.5 ratio is deduced.

On-demand EC2 Instances vs Dedicated Servers

SPOT INSTANCES

Spot instances enable you to request spare computing capacity from AWS when they have it and are offered at up to 90% off the on-demand price. However, they’re not available all of the time so can’t be used in isolation to run a business.

The current generation of general purpose spot instances are priced in the table adjacent, with a comparison to prebuilt servers. A prebuilt server is a physical server that is pre-racked and has a specification that cannot be altered.

These servers are well suited to tasks that require rapid provisioning, for temporary environments, perhaps to cater for short term, but highly significant, bursts in demand.

Spot Instances vs Prebuilt Servers

RESERVED INSTANCES

AWS Reserved Instances are virtual servers that run in Amazon Web Services’ Elastic Compute Cloud (EC2) and Relational Database Service (RDS). Instances are purchased at contract prices, plus hourly rates. Instances are offered in various levels of compute power and there are two different purchase options:

  1. Standard Reserved Instances
    These provide you with a discount of up to 75% compared to on-demand instance pricing, and can be purchased for a 1-year or 3-year term. Customers have the flexibility to change the Availability Zone, the instance size, and networking type of their Standard Reserved Instances.
  2. Convertible Reserved Instances
    These provide additional flexibility, such as the ability to use different instance families, operating systems, or tenancies over the Reserved Instance term. Convertible Reserved Instances provide you with a discount of up to 54% compared to on-demand instances and can be purchased for a 1-year or 3-year term.

Price comparisons will be made with convertible reserved instances, as the additional flexibility they offer is also available with dedicated servers.

The AWS Reserved Instances “All Upfront” option is akin to the purchase of dedicated servers, thus its prices have been used in the table below.

As the name suggests, customers pay for the entire Reserved Instance term with one upfront payment, making it the cheapest choice. For the purposes of comparison, the upfront cost has been divided into monthly portions.

For RapidSwitch, the discounts for committing to a contract instead of on-demand are 15% for one year and 40% for three years.

Convertible Reserved Instances vs Fixed Term Dedicated Servers

DEDICATED HOSTS

Dedicated hosts are effectively dedicated servers. However, due to the premium charged by AWS for each instance, the cost comparison against traditional servers highlights just how expensive this model is.

The RAM calculation is an approximation based on the amount of instances that can be fit into a dedicated instance, as per the information here and here. GPU hosts have been excluded, as it is not possible to determine the number within a dedicated host, and, unless unavailable, the location used is London.

Dedicated Hosts vs Dedicated Servers

Resources: Storage

AWS storage is divided into four different categories:

1. AMAZON ELASTIC BLOCK STORAGE (AMAZON EBS)

Amazon EBS is persistent local storage for Amazon EC2, for relational and NoSQL databases, data warehousing, enterprise applications, Big Data processing, or backup and recovery. It is designed for workloads that require persistent storage, typically associated with large data warehouses. There are five different EBS volumes:

  • Amazon EBS General Purpose SSD (gp2) volumes
  • Amazon EBS Provisioned IOPS SSD (io1) volumes
  • Amazon EBS Throughput Optimized HDD (st1) volumes
  • Amazon EBS Cold HDD (sc1) volumes
  • Amazon EBS Snapshots to Amazon S3

For the purposes of comparison, 100GB of general purpose SSD will be used. General purpose SSD dictates that storage is charged by the amount you provision in GB per month until you release the storage. Provisioned storage for gp2 volumes is billed in per-second increments, with a 60 second minimum. I/O is included in the price of the volumes, so you pay only for each GB of storage you provision.

With a traditional infrastructure model, storage would be provided by hard drives in the servers themselves or through connections to shared storage devices, such as SANs. For 100GB of storage, the price is inclusive of the server prices above.

Amazon Elastic Block Storage vs Traditional Server

2. AMAZON ELASTIC FILE SYSTEM (AMAZON EFS)

This provides a file system interface and file system access semantics to make data available to one or more EC2 instances, for content serving, enterprise applications, media processing workflows, Big Data storage, or backup and recovery. It’s a premium service designed to auto-scale with capacity.

The storage amount billed in a month is based on the average storage space used throughout that month. Storage usage is measured in “GB-Months,” which are added up at the end of the month to generate monthly charges.

For example, take a file system located in Ireland, using 600 GB of storage for the first 12 days in April and 500 GB for the last 19 days in April.

At the end of April, the usage in GB-Hours would be:

Total usage (GB-Hours) = [600 GB x 12 days x (24 hours / day)] + [500 GB x 19 days x (24 hours / day)] = 400,800 GB-Hours

GB-Hours are added up and converted to GB-Months to calculate monthly charges. In Ireland, one GB-month costs £0.2445.

Total monthly storage charge for 1100GB usage in April = 539 GB-Months x £0.2445 = £131.79

Amazon EFS is equivalent to an additional server running as a NAS. RapidSwitch can provide bespoke NAS solutions, but, as one example, a server with 8x 2TB in RAID5 provides 14TB of useable space and is available for £120 per month, which is less than 1p per GB.

Another option is to simply take raw NAS space from RapidSwitch’s shared NAS environments, priced at a flat 2.5p per GB.

3. AMAZON SIMPLE STORAGE SERVICE (AMAZON S3)

A scalable, durable platform to make data accessible from any internet location, for user-generated content, active archive, serverless computing, Big Data storage, or backup and recovery. S3 is the standard choice for many businesses and individuals. Customers create a storage account, choose a region to place it, use it, and are billed for only what’s used.

There is no direct equivalent to S3 in the traditional model for an organisation to procure on a ‘pay-for-use’ model. For an organisation to build their own version of S3, it would involve a significant amount of investment in infrastructure and be a different cost model.

As with any AWS product, users should be mindful of the fees for transmitting data to and from storage locations as the transit costs can become prohibitively expensive.

Amazon S3 Storage Costs

4. AMAZON GLACIER

This is highly affordable long-term storage that can replace tape for archive and regulatory compliance. Glacier is a solution for customers who want low-cost storage for infrequently accessed data but is not appropriate for running a business with.

There is no minimum fee with Amazon Glacier – users only pay for what they use.

Similar to S3, although service providers will have platforms for archive and storage available to their customers, this is outside of the scope of this guide. An organisation could build their own archive storage platform but it would not be on the same procurement model as AWS Glacier.

Again, the speed and costs of retrieving data from Glacier mean that users should run through a series of business scenarios prior to choosing any storage product to ensure the product’s features and costs are matched to the correct business model.

Resources: Databases

AWS provide managed relational, non-relational, data warehouse, in-memory data store, and graph databases. These database instances will also require storage, whereas a traditional server will incorporate the storage or be connected to a shared storage device.

One such managed database is Amazon RDS, which enables users to set up, operate, and scale a relational database in the cloud. Customers have a choice of six database engines: Oracle, Amazon Aurora, PostgreSQL, Microsoft SQL Server, MariaDB, and MySQL.

Direct comparisons can therefore be made with Amazon RDS running MySQL and a dedicated server doing the same. While paid and unpaid editions of MySQL are available, Amazon RDS for MySQL doesn’t come “bundled” with a paid edition – their website states the free MySQL Community Edition versions 5.5, 5.6 and 5.7 are supported. Therefore, the potential cost of MySQL doesn’t need to be included when making comparisons.

Similarly, bandwidth has been excluded since bandwidth between instances is free, but it should be remembered that bandwidth leaving AWS for users pulling data will be charged accordingly.

Amazon RDS vs Dedicated Server

With Amazon RDS Multi-AZ deployments, Amazon RDS automatically creates a primary DB instance and, at the same time, replicates the data to a standby instance in a different Availability Zone (AZ). This enhances the durability and availability of database instances.

Amazon doubles the cost for these deployments, while with RapidSwitch, it is multiplied by 2.5 – for doubling the amount of servers required and including the extra cost for networking between them. This is a good example of the benefits of the public cloud – a simple check box adds resilience, whereas expanding infrastructure will add to the complexities of managing the environment.

Amazon RDS Multi-AZ Deployment vs Dedicated Server Equivalent

Data Transfer/Bandwidth

On-demand Data Transfer from AWS lets you pay for data transfer by the GB with no long-term commitments. All data transfer in is either free or provided at a minimal cost, but the costs for transferring data out can quickly add up.

Up to 10TB/month of data transfer out from Amazon EC2 to the internet costs £0.06668 per GB, with the first GB free. So, a business using 10 TB in a month would spend £666.80.

However, in stark contrast, almost all dedicated server providers bundle in 10TB of bandwidth for free. The upshot of this is that for heavy users of bandwidth, AWS simply isn’t cost-effective.

When is AWS the best fit?

AWS is a strong contender when a small amount of a very complex service – that is multi-region and high-resilience – is needed. In these cases, you’re unlikely to recover your investment by building your own platform. What’s more, the cost and expertise needed to design it would be considerable – while AWS has the product base to be able to sell it many times over and the expertise on-hand to develop the best tools.

HAPPI: AN AWS USE CASE

Happi is an excellent example of an application that’s a perfect fit for AWS. We chose AWS as the hosting platform for Happi because the application was being built from the ground up, native to AWS. Happi will never require a large footprint of infrastructure, no matter how many users are active. It is effectively a text-based catalogue of services and some lightweight images, with API calls made to service provider platforms for integration.

Further reasons we chose AWS were:

  • Bandwidth: As stated, Happi consumes a low amount as the images are small and environments are described using JSON. As we’ve shown, bandwidth can be very costly in AWS for big consumers but is manageable for light users.
  • Serverless technology: This was perfect for us as it allowed Happi to be easily modelled and deployed while our software developers could focus on improving application usability and management. There are no instances running in AWS to support Happi.
  • Global reach: The mission for Happi is to provide users with the ability to deploy products and services into any location in the world. Using AWS’s network, Happi is able to easily scale to any global region, providing worldwide coverage.

AWS Support

This guide has highlighted the comparable costs of infrastructure versus AWS instances and shown that costs are considerably higher on AWS instances. The higher costs of AWS are sometimes justified by the notion that their customers no longer require support, but that is a fallacy. Support is required, just in a different form.

The first question to ask is, “will we cut all IT staff headcount in moving to AWS?”. Often the answer is ‘no’, as IT staff support internal users. In addition to supporting internal users, they will liaise with the application vendors for app support and fixes and this is done alongside the maintenance of the infrastructure environment that the app sits on. If the applications are all moved to AWS, the argument is that the maintenance requirement goes too – except it doesn’t. The environment still needs to be monitored to ensure it is up, that services are running and, so, will require an additional skill set from staff to learn how AWS works.

If traditional servers are lifted and move to AWS instances, the same level of monitoring, management and support is required for those instances.

This is why any true migration to the public cloud requires a re-development of the application so that it is ‘cloud native’. This will see the reliance on AWS instances drop considerably, and proprietary AWS services will be utilised instead. However, this is a completely different proposition. The support for such an environment sits with the application developers as the software is intertwined with the hosted environment.

If an application vendor allows an organisation to self-install their application on to their own AWS environment, or if the application is developed in-house, the traditional in-house IT staff will need to have a strong DevOps capability to understand how the AWS environment links with the application or the in-house developers will have to start providing support for end-users.

The upshot is that staff are still necessary, they just perform a different role – employees have to start doing things the AWS way. The AWS approach can of course be learnt, with role-based learning and certification programs available. At foundational, associate, and then professional level, employees can become AWS certified in these areas:

  • Cloud Practitioner
  • Solutions Architect
  • Developer
  • DevOps Engineer
  • SysOps Administrator
  • Advanced Networking
  • Big Data
  • Security

But the point is that AWS adoption is not as lightweight as often thought and, in times of need, where assistance and support are required, AWS often falls short. Companies are routinely forced to introduce a third-party advisor into the mix which, predictably, adds to the costs.

Conclusion

AWS is a market leader for good reason and a perfect hosting platform for applications such as Happi. However, the notion that it is always cheaper and never requires support is a myth. AWS (and other big public cloud providers) are not, and never will be, the best fit for all use cases.

We built Happi to level the playing field and become a marketplace for finding and comparing hosting solutions. Users are able to compare prices for both physical and virtual infrastructure from big and small providers alike and then pick the right tool for the job.

As the world’s first true IaaS tool, Happi lets users design, compare, buy, and deploy hosted infrastructure anywhere in the world from global vendors, such as AWS. No longer does public cloud have to be the go-to default – with Happi, it’s easy to see how the likes of AWS stack up against private offerings from all across the globe.

Get Happi

Happi is ready and waiting to revolutionise your infrastructure provisioning.

Sign-up and start using the tool for free now.