Professional Documents
Culture Documents
Cloud Pricing Report 2021 22.07.21
Cloud Pricing Report 2021 22.07.21
Key findings
• The reference cloud pricing from leading public cloud providers is
very similar.
• The reference cloud pricing from leading private cloud providers is
very divergent.
• Public clouds continue to be the most cost-efficient platform for
small-scale deployments.
• Private clouds continue to be the most cost-efficient platform for
medium-scale and large-scale deployments.
• A multi-cloud architecture emerges to be the most economical cloud
architecture, ensuring maximum CapEx and OpEx efficiency.
Introduction
Estimating and comparing cloud pricing between public and private clouds has
always been challenging. While leading public cloud providers maintain the
official pricing for their cloud services in a form of cost-per-resource metrics,
resembling the same in the private cloud space is not trivial. As a result, the
most effective way to compare costs between public and private clouds is to
estimate the total cost of ownership (TCO) of the workloads and use it as a
subject of the comparison.
In the following report, we present the reference cloud pricing from leading
public and private cloud providers as of June 2021. We compare sample cost-per-
resource metrics for Amazon Web Services (AWS), Azure and Google Cloud
Platform (GCP) instances based on the same criteria. In turn, on the private cloud
front, we compare license and delivery costs, and support service fees for
VMware, Red Hat OpenStack Platform and Canonical’s Charmed OpenStack. We
also demonstrate how using TCO calculators enables organisations to make the
right decision regarding their cloud architecture based on economic criteria.
Reference cloud pricing
The reference cloud pricing is what organisations usually check in the first place
when trying to estimate the TCO of their workloads. Since all presented
platforms provide the same level of enterprise support, in most cases the
decision of which cloud infrastructure to choose is driven by economics. While
public clouds prove to be more economical for small-scale deployments, private
clouds are more cost-efficient at a larger scale. Therefore, using a multi-cloud
architecture, consisting of the public cloud infrastructure and cost-effective
private cloud is usually the most optimal solution.
In order to compare cost-per-instance metrics across AWS, Azure and GCP the
following criteria have been applied:
The reference pricing from leading public cloud providers is shown in Tab. 1.
Instance type AWS instance / price Azure instance / price GCP instance / price
n2d-standard-4 /
General-purpose m5a.xlarge / $0.172000 D4a v4 / $0.192000
$0.190320
n2d-highcpu-4 /
Compute-optimised c5a.xlarge / $0.154000 F4s v2 / $0.169000
$0.140508
n2d-highmem-4 /
Memory-optimised r5a.xlarge / $0.226000 E4a v4 / $0.252000
$0.256736
2
Based on the presented data it can be seen that AWS provides the lowest
reference pricing for general-purpose and memory-optimised instances.
Although GCP provides the lowest reference pricing for compute-optimised
instances, it is important to mention that such instances on GCP use 1 GB of
random-access memory (RAM) per 1 vCPU, while the same instance types on AWS
and Azure use 2 GB of RAM per 1 vCPU. This allows Google to offer compute-
optimised instances at a better price. Besides this, there are no significant
differences in the reference pricing from leading public cloud providers.
The reference pricing from leading private cloud providers is shown in Tab. 2.
Support and
Not available Not available $5,475 per host
fully-managed
* - Red Hat OpenStack Platform reference pricing is not publicly available. The
provided prices are indicative and are based on third-party reports and sources.
Based on the presented data it can be seen that the reference pricing and the
pricing model differs significantly across leading private cloud providers. First of
all, VMware is the only vendor that requires licenses to be purchased for their
vRealize platform. In turn, while both VMware and Canonical offer consulting
services at a fixed price, Red Hat charges based on the duration of the
engagement - which is generally between several weeks and months. When it
comes to the support subscription VMware and Red Hat make them dependent
on the number of CPUs or VMs, while Canonical simply charges per the number
of hosts. Canonical is also the only vendor that offers fully-managed services for
their Charmed OpenStack platform.
3
TCO calculators
Although the reference cloud pricing helps to roughly estimate costs, in reality, a
lot of other factors have an impact on the total cost of ownership (TCO). This is
because additional charges are applied based on the storage and network
consumption, significant discounts are available when running workloads long-
term, or in the case of private clouds, one should include hardware costs, hosting
charges and operations and maintenance costs in their estimates. Therefore,
when estimating the TCO it is best to use TCO calculators. The following section
provides a brief overview of the most popular ones.
4
Sample cost scenarios
In the following section, we demonstrate sample scenarios and provide guidance
on the cloud architecture choices. All estimates were obtained using TCO
calculators presented in the “TCO calculators” section, augmented with the
reference pricing presented in the “Reference cloud pricing” section wherever
needed. All public cloud estimates assume a reserved instances pricing model
with a 3-year timeframe and Ubuntu guest OS. All private cloud estimates
assume a 3-year hardware renewal period and include reference hardware costs,
typical hosting charges and operations and maintenance costs. It has been
assumed that the average annual operations team staff salary is $125,000 and
the operations team size is 12 for 24/7 maintenance with 99.99% uptime.
Amount of
Number of Amount of Network Additional
Purpose persistent
vCPUs RAM [TB] requirements information
storage [TB]
Database 24 0.2 2 No No
The estimated TCO for this system across leading public and private cloud
providers is shown in Tab. 4 and Fig. 1.
Charmed OpenStack
637,500 143,700 1,068,600
(fully-managed)
5
Fig. 1. Small-scale TCO estimates.
Amount of
Number of Amount of Network Additional
Purpose persistent
vCPUs RAM [TB] requirements information
storage [TB]
Moderate
Database 500 4 40 No
inter-AZ traffic
Used during
Developer VM 200 0.8 5 No business
hours only
ARM CPUs
required; Used
Developer VM 20 0.08 0.5 No
during business
hours only
6
The estimated TCO for this system across leading public and private cloud
providers is shown in Tab. 6 and Fig. 2.
Charmed OpenStack
637,500 143,700 1,068,600
(fully-managed)
7
Scenario 3: Large-scale - Video streaming system
In this scenario, an entertainment company is hosting their own video streaming
system. The system consists of a data warehouse, data lake, data analytics, video
transcoding engine and web application. The roughly estimated requirements
for this system are shown in Tab. 7.
Amount of
Number of Amount of Network Additional
Purpose persistent
vCPUs RAM [TB] requirements information
storage [TB]
Moderate
Data warehouse 32,000 256 5,120 No
inter-AZ traffic
Run daily
Data analytics 12,000 24 0 No
for 8 hours
Video
Daily 8-hour
transcoding 24,000 48 0 No
load spike
engine
The estimated TCO for this system across leading public and private cloud
providers is shown in Tab. 8 and Fig. 3.
Charmed OpenStack
13,387,000 3,577,000 24,113,000
(self-managed)
8
Fig. 3. Large-scale TCO estimates.
9
Executive summary
Analysing cloud costs is a non-trivial process that requires understanding the
workloads running in the cloud as well as using proper tools to compare costs
across various cloud providers. While the reference pricing from leading public
cloud providers is very similar, the reference pricing in the private cloud space
differs substantially across leading vendors.
While it is not possible to directly reference cloud pricing between public and
private cloud providers, using TCO calculators enables estimating the TCO of the
workloads and making the decision regarding workloads placement based on
economics criteria. Although public clouds continue to be the most cost-efficient
platform for small-scale deployments, using private clouds is more economical at
medium and large scale. Moreover, leveraging a multi-cloud architecture proves
to be the most optimal choice from an economic standpoint.
© Canonical Limited 2021. Ubuntu, Kubuntu, Canonical and their associated logos are the registered trademarks
of Canonical Ltd. All other trademarks are the properties of their respective owners. Any information referred
to in this document may change without notice and Canonical will not be held responsible for any such changes.
Canonical Limited, Registered in England and Wales, Company number 110334C Registered Office:
12-14 Finch Road, Douglas, Isle of Man, IM99 1TT VAT Registration: GB 003 2322 47