Reliable Cloud Performance
People are right to be wary of any vendor claiming to be the “top performing!” or “fastest!” cloud provider. Most folks know that ANYTHING can look spectacular – or unspectacular for that matter – if you stack the deck just right. But at the same time, cloud shoppers have a deep hunger for legitimate information on realistic performance expectations. Cloud performance has a direct impact on what you spend on compute resources, how you decide the right host for your workload, and how you choose to scale when the need arises. In this blog post, we’ll summarize some recent findings and put them in context.
With the launch of our new Hyperscale instances, we at CenturyLink Technology Solutions approached an independent analytics company, CloudHarmony, and asked them to conduct an extended performance test that compared CenturyLink Cloud Hyperscale servers to the very best equivalent servers offered by AWS and Rackspace. CloudHarmony is a well-respected shop that collects data from dozens of benchmarks and shares the results publicly for anyone to dissect. After running a variety of benchmarks over a long period of time (to ensure that the test gave an accurate look over an extended window), they shared their findings with the world.
The results were positive – as we’ll talk through below – but how do reliable performance metrics help you in your cloud journey?
More Bang for the Buck
In an ideal world, you want reliable performance at a fair market price and no hidden charges. In the CloudHarmony results, we saw that our Hyperscale SSD storage provided excellent disk read performance and strong disk write performance through a variety of tests. In the results below – run against AWS c3 servers and Rackspace Performance servers – you can see that Hyperscale has a fantastic IO profile for large block sizes.
Disk Read Performance
Why does this matter? Consider databases running on Microsoft SQL Server that often works with 64k blocks. By running this workload on Hyperscale, you get persistent storage, high performance, and no charges for IO requests or provisioned IOPS. This results in predictable costs and fewer resources needed to achieve optimal performance.
Simplified Decision Making
Choice is great, but is also a paradox. When you’re faced with dozens of server types to choose from, you find yourself selecting a “best fit” that may compromise in one area (“too much RAM!”) in order to get another (“need 8 CPUs”). In CenturyLink Cloud, we have two classes of servers (Standard and Hyperscale) and both have shown to have reliable performance. Pick whatever amount of CPU or memory that makes sense – which is of course how traditional servers have always been purchased.
Choose Your Own VM Size
If built-in data redundancy doesn’t matter, but reliable, high performance does, choose Hyperscale. Need strong, consistent performance but want daily storage snapshots and a SAN backbone? Use Standard servers. Straightforward choices means that you spend less time navigating a gauntlet of server types and more time deploying killer applications.
Predictable Performance & Scaling
Valid performance testing results can help you understand how best to scale an application. Should I add more capacity to this VM, or does it make sense to add more VMs to the environment? That’s a hard question to answer without understanding how the platform reacts to capacity changes. The CloudHarmony results not only showed that the CenturyLink Cloud Hyperscale CPU performed better than the others in the “Performance Summary Metric” that compared cloud servers to a bare metal reference system, but also showed that performance improved as CPU cores were added. That’s obviously not shocking, but it’s good to see that performance change was relatively linear.
How does this information help you maximize your cloud portfolio? If you know that you can add resources to a running VM *before* scaling out to new hardware, that can simplify your infrastructure and lower your costs. Scaling out is fantastic cloud pattern, but it doesn’t always have to be the first response. You can trust that Hyperscale scales out *and* up well, and you can plan your scaling events accordingly.
Performance metrics are only a snapshot in time. The individual results may change from month to month or year to year, but a reliable performance profile means that you can minimize costs, make decisions faster, and make predictable choices.
By Richard Seroter