It seems like everyone is jumping on the cloud bandwagon these days. Cloud software-as-a-service (SaaS) solutions are the go-to answer for many companies looking to offload infrastructure and management needs for email, customer-relationship-management (CRM) software, and other common apps. An increasing number of companies are also turning to cloud deployments for real-time or near-real-time database processing. By moving processing and analytical services to the cloud, companies can reduce their infrastructure costs and accelerate performance for online transaction processing (OLTP) or online analytical processing (OLAP), depending on the hosting provider and offering.
No Two Clouds Are Alike
Businesses often use virtual-server environments for infrastructure-as-a-service (IaaS) cloud deployments. Providers like Amazon Web Services (AWS), Microsoft Azure, Google Cloud Platform, Rackspace, and others generally prefer using virtual machines because they can be easier and cheaper for vendors to deploy and manage compared to physical—or bare-metal—servers. Despite their prevalence, virtual servers aren’t necessarily ideal for businesses with high performance needs; for these businesses, bare-metal servers might be a better choice because physical machines don’t have the overhead of a hypervisor or shared tenancy.
So how do bare-metal servers and virtual machines compare in real-world scenarios? Is there really a measureable difference in performance? To find out, we decided to run some comparison tests in our labs at Prowess. In our testing, we aimed to determine which IaaS cloud deployment—bare metal or virtual—offered the best performance under load.
Putting Cloud Deployments to the Test: Bare-Metal vs. Virtual Servers
To create a real-world test scenario, we compared the performance of identical workloads on SQL databases for two self-service IaaS implementations:
- Rackspace OnMetal I/O v2 bare-metal cloud server
- Amazon Elastic Compute Cloud (Amazon EC2) Type i2.4xlarge virtual-server instance
We measured the input/output (I/O) performance for each configuration as we increased the number of virtual users to ramp up the load on the databases. Our results were definitive: the Rackspace bare-metal cloud deployment handled more than double the peak transactions per minute (TPM) of the Amazon EC2 virtual deployment (see Figure 1).
Figure 1. Transactions per minute (TPM) for Rackspace OnMetal I/O v2 (red) versus Amazon Elastic Compute Cloud (Amazon EC2) (blue)
In addition, the OnMetal I/O v2 implementation scaled faster than the Amazon EC2 deployment, which means that businesses running OnMetal I/O v2 can accommodate more users with the same or fewer instances.
Because our tests verified TPM without any database-specific tuning, the results suggest that users can also expect real-world performance gains for almost any high-I/O use case involving large caches, search indexes, or in-memory analytics, such as payments processing, enterprise resource planning (ERP), inventory control, operations, or finance.
To get the full story on our testing and results, read the test report on our website.