Part 2: Taking a closer look at how on-prem infrastructure compares to Amazon Web Services (AWS) as we unpack the details around the comparison of the “all-in” TCO for each of the tested configurations.
In Part 1 of this series, we provided an overview of the results from the HPE on-prem vs. AWS study and a brief primer on AWS. Here in Part 2, we dig deeply into the total cost of ownership (TCO) for each of the solutions evaluated in the study. You can download the complete technical paper here: HPE On-Prem vs. AWS. (Registration is required.)
Embarking on a comparison of on-premises vs. public cloud IaaS
At the time we began this comparative journey, we had already optimized an HPE on-prem solution configuration to deliver the best performance while keeping an eye on the cost of the configuration. We evaluated price-performance trade-offs of various processor SKUs, memory sizes, disk configuration alternatives and networking options to build a configuration that produced the best performance at the lowest cost. Once we settled on a good balanced solution, we turned our attention to the AWS configuration.
We leveraged the knowledge gained through sensitivity analysis in the on-prem components to select the best AWS solution, but also considered other architectural alternatives offered by AWS, such as all SSD and EBS (Elastic Block Store) Network accessible storage. At that time, AWS had not yet released an instance for the new Skylake (Intel Scalable Processor Family) CPU, so we initially focused only on the Broadwell (E5 generation) of Intel Xeon processors. As we neared the end of the evaluation, the AWS m5 instance became available containing the newer Skylake generation so we repeated the comparison with optimized configurations including this latest Intel generation.
Equivalent configurations
Our goal was to make every comparison in the study as apples-to-apples as possible matching attributes of compute, memory, storage and networking as closely as possible. In some cases, we could not make an exact match of characteristics such as core count, frequency or bandwidth, but we got as close as the available AWS instances would allow.
Each configuration has nine worker nodes (HPE) or instances (AWS) and three management VMs or instances. At the time of the publication of the whitepaper, the only Skylake instance available from AWS was the m5 instance, which uses EBS exclusively for storage. For the older Broadwell generation, both EBS and all-SSD instances were available, so we continued to include the Broadwell comparisons in the study to provide a comparison of the effects of these options with the workload. Note that AWS did not provide instances that allow a combination of both EBS and SSD storage, so we were limited to one or the other. We chose to include two Broadwell AWS instance types to add these comparison points: the i3, where the storage is all-SSD, and the m4, where the storage is all EBS.
The worker nodes/instances do most of the work while the management nodes/instances coordinate and deploy the workloads on the worker nodes/instances.
The configurations: A closer look
Table 1 shows more detail for the CPU SKUs, memory, storage and networking configurations.
As we described in the first blog of this series, we used the Amazon Simple Monthly Calculator to derive the TCO for each of the AWS configurations across several purchase options:
- 3-Yr, paid all upfront, reserved instances
- 1-Yr, paid all upfront, reserved instances
- On-demand instances
For the on-demand instances, all payments are monthly payments with no upfront one-time payment. For the reserved instances, the instance portion of the bill is paid all upfront in a one-time payment and the portion for the associated support costs and EBS volumes is billed monthly. For comparison, we amortized the on-time payment(s) over 36 months and 12 months for the 3-Yr and 1-Yr purchase options, respectively. We then combined this with the ongoing monthly costs to generate a single monthly TCO. We did not account for any interest on the amortization in either the AWS Reserved Instances nor the amortization of the HPE on-prem configurations.
Appendix B of the On-Prem vs. AWS technical paper shows a representative session of the Simple Monthly Calculator for the m5 (Skylake) 3-Yr, paid all upfront, reserved instance configuration. As a fun exercise, the reader can fairly quickly price out one of the AWS configurations in this study and apply the configuration to the Amazon Simple Monthly Calculator. If you do this, you will come up with a higher cost for the m4 and m5 configurations than we show. This is due to the unexpected high cost of the IOPS provisioned volumes that we used and we did not want to increase the TCO calculation of the AWS so substantially when it wasn’t architecturally necessary. (This is discussed in Appendix E of the technical paper and we will dive deeper into the topic in Part 4 of this series.)
TCO calculations: HPE on-prem
The TCO for the HPE on-prem configurations are the all-in costs to operate the infrastructure, including the hardware, software, labor, power, maintenance, datacenter infrastructure (power generation capital equipment) and space infrastructure. The data center and space infrastructure accounts for 2x over-provisioning to account for future growth. Refer to Tables 2 and 3 in the On-Prem vs. AWS technical paper for details of these costs and Appendix C for the fully costed bill of materials (BOMs) for both the HPE Gen9 and Gen10 server configurations. For the on-prem configurations, the additional costs we calculated for power and space infrastructure, labor, maintenance, power consumption and carbon footprint adds 37% over the cost for just the hardware and software. For the AWS solution, these costs are baked into the price of the services. As with the AWS reserved instances, we did not account for any interest charges on the amortized amounts. Please refer to the On-Prem vs. AWS technical paper for additional cost break-down of the configurations.
TCO comparisons
Here, Figure 1 shows the monthly TCO for the Broadwell and Skylake configurations used in this study. The figure shows the 3-Yr amortized monthly costs for the HPE on-prem configurations and each of the AWS configurations with the 3-Yr, paid all upfront, reserved purchase option. This purchase option from AWS is the lowest cost option. For the Skylake comparisons, the AWS monthly TCO is 1.7 times the cost of the TCO for the HPE on-prem solution.
The on-demand purchase option requires no upfront payments and no commitments. It is convenient to start up and tear down with a pay-per-use billing model. However, these benefits do come at a stiff premium. The AWS monthly premium for the on-demand purchase option for the m5 configuration is 2x more than the cost of the reserved purchase option and 3.4 times the cost of the TCO for the HPE on-prem solution.
TCO comparisons of the configurations are one piece of the equation and the performance of each configuration is another.
Coming up next
In Part 3 of our series, we step back and ask Dr. Paul Cao, master technologist, to dive deeply into the cloud-scale advanced analytics workload that we used for this study and to discuss the measured performance results for each of these configurations. Please provide any comments, questions or feedback on the series so far.
Follow the blog series
In this blog series, we present details and insights around the HPE and AWS comparison for the following topics:
Part 1: Introducing the HPE on-prem vs. AWS primer
- An overview of the study and the summary of findings in the comparison of HPE and AWS
- An AWS primer to provide a brief overview of what is available in AWS EC2 IaaS capabilities
Part 2: Total Cost of Ownership (TCO)
- The configuration options and selected configurations: HPE and AWS
- The “all-in” cost analysis for the on-prem configuration, including costs for maintenance labor, data center infrastructure, energy and cooling, carbon footprint and warranty
- The cost analysis for the AWS configurations based on reasonable configurations and purchase options
Part 3: The workload and throughput (performance)
- Overview of the cloud-scale advanced analytics workload
- The throughput measurements for each configuration in total Queries per Minute (Qpm)
- Analysis of each architecture and comparison of Local SSDs and EBS
Part 4: Price performance and control
- Price-performance for each configuration
- A look at the cost of high-throughput EBS volumes
- Analysis of ability to control attributes of performance, data sovereignty, privacy, security
- Pay per use with variable payments based on actual metered usage
- Dynamic and instant growth flexibility
- Onsite extra capacity buffer
Meet Infrastructure Insights blogger Lou Gagliardi, Sr. Lab Director, Enterprise Solutions & Performance, Hybrid IT, HPE.
Infrastructure Insights
Hewlett Packard Enterprise
TechExperts
Our team of HPE and other technology experts shares insights about relevant topics related to artificial intelligence, data analytics, IoT, and telco.
