byvaleriavong09-27-201609:04 AM - edited 11-01-201612:03 PM
On the Solutions & Performance team at Nutanix, we’re often working with customers or prospective customers to help them identify what they really need their hyperconverged infrastructure to do and how to evaluate their platform options based on those specific requirements. There are several benchmarking tools available to help us figure out what a block is doing—and what it could do. When configured properly, each tool shows different aspects of a system’s general performance characteristics in certain scenarios.
There’s no such thing as a “one size fits all” performance testing tool. In an ideal situation, you’d want to assess your HCI with actual application workloads rather than by simply generating I/O. Datacenters in real production environments need to provide resiliency and fault isolation, data availability, and consistency as well as fast I/O management.
Adding tools such as HammerDB and Benchmark Factory (which generate real database transactions and simulate user activity) or LoginVSI (which performs real desktop work with various applications) to your testing kit can help fill out the picture of how a system would behave under everyday conditions. Of course, the more realistic a test is, the more complicated and time-consuming it can be to set up.
With that in mind, sometimes you just need to run Iometer, and that’s fine. Iometer is a convenient tool that uses random data to mimic I/O patterns and sizes to simulate workloads putting stress on storage. As it’s a fairly general-purpose tool, you’ll want to have specific criteria in mind before you begin using Iometer, so you’ll know what the results are telling you.
If you’re running Iometer, here are some tips for getting the most representative test results and thinking about them in a real-world context.
Make sure you’re testing the best node type for your workload requirements. Nutanix offers several different node configurations to meet various customer needs.
Update your Nutanix software. Acropolis gets better with every release, and running an AOS version below 4.6 doesn’t give you realistic results.
Turn data integrity protections on. Nutanix has checksums enabled at all times and compares blocks on read, because even the fastest system in the world is worthless if it loses your data. Are all the systems you’re testing keeping your data safe?
Run Iometer with unformatteddisks. Using raw disks directly tests the storage controller, so you’re not inadvertently also testing the guest OS filesystem and caching. This approach helps to keep results consistent.
Test a big enough working set. Reading a small dataset from RAM isn’t a valid measure of performance for applications that exceed the size of the system cache. Few real applications generate I/O sizes of under 4 KB; databases tend to see 8 KB I/O and up.
Use sustained write tests to look for consistency and predictability in performance over time. Customers running database applications and other critical systems always want to avoid wild variations, which can have a major impact on operational costs and SLAs. To test for stability even under high workloads, try using 512 KB.
Run tests using random I/O. Most I/O profiles are random in nature, especially as you combine multiple workloads and run them side by side. Testing with random I/O lets you see how well the system provides quality of service and deals with noisy neighbors—aspects of performance that contribute to reliability.
Keep Iometer’s limitations in mind. While Iometer is great for testing a system’s limits and for understanding how performance responds under different controlled conditions, no actual application performs only one size or type of I/O, and no real production environment runs a steady stream of I/O at full blast all the time.
In whatever test suite you use, be sure to look for network bottlenecks as you scale up the test environment. By preserving data locality, Nutanix keeps its network available for service application transactions and ready to respond during maintenance and failure scenarios. Network availability is especially important with newer all-flash technologies.
Comparing platform options can be a daunting task, and test results from different vendors and tools can vary widely. We hope that these Iometer best practice tips can help empower you to get the most benefit out of your comparison performance testing.
This post was authored by Kate Guillemette, Technical Writer at Nutanix, Will Strickland, Solutions & Performance Engineer at Nutanix, and Michael Webster, Manager, Solutions & Performance Engineering at Nutanix.
Disclaimer: This blog contains links to external websites that are not part of Nutanix.com. Nutanix does not control these sites, and disclaims all responsibility for the content or accuracy of any external site. Our decision to link to an external site should not be considered an endorsement of any content on such site.
Nutanix Xpress is a new solution designed to bring the power of the Nutanix enterprise cloud platform to small and medium businesses. Nutanix Xpress is so simple that a single IT professional can install and manage an organization’s entire IT infrastructure, freeing valuable time to focus on initiatives that directly impact the top line.
Our team put together a series of three videos to demonstrate how easy it is to get started with Nutanix Xpress. Enjoy!