In the first cloud benchmark of its kind, 75,000 virtual servers were spun up on a Mirantis OpenStack cloud of 350 physical servers spanning multiple data centers, proving that it’s possible to build very large clouds on OpenStack quickly and reliably OpenStack could respond to on-demand, real-world workloads.
This test, spanning eight hours, was made possible at this scale thanks to a partnership with IBM’s cloud division SoftLayer, and “Rally” an open-source OpenStack testing and benchmarking tool backed by Mirantis, a company that specializes in software, services, training and support for OpenStack.
Mirantis president and CEO Adrian Ionel said this benchmark is significant because it will help large businesses understand that OpenStack can be used to quickly provision and effectively manage a large number of VMs across multiple data centers.
“We’re pushing OpenStack to its limits in harsh conditions that mimic real world production workloads and demanding customer use cases,” he said. “The benchmark is incredibly exciting and relevant because it shows a number of things.”
Servers in this OpenStack cloud were as far 1,800 miles away from each other, yet the cloud achieved very low latency, behaving as though they were located in one data center.
“It’s important to understand the setup, it’s not just a one-data center setup; it’s a multi-data center cloud in data centers very far apart,” Ionel said. This is a crucial feature for many enterprises that already have infrastructure in different locations, and that want the advantages of resiliency in case one of the data centers goes down.
It was also found that new resources could be provisioned very quickly.
“Not only can we build a very large cloud on OpenStack, but we can build one that is very fast, very responsive, which is also very important,” Ionel said. “Enterprises have a large number of concurrent users who constantly demand new resources from the cloud.” The benchmark show that the cloud could deliver one new VM to end-users in a fraction of a second.
Mirantis was also able to glean some information on how to tweak and tune the Mirantis OpenStack architecture but also the other software and technologies used to build a complete cloud such as HAProxy for load balancing.
“What we learned is how to do is to tune all these things, and how to setup the parameters properly, including the replicated multi-master database, which can be a key bottleneck in benchmarks like this,” he said.
While this benchmark was used to test Mirantis OpenStack, the testing and benchmarking tool it used, Rally, is available to OpenStack cloud providers and companies hoping to test and optimize their private clouds.
“Rally can be used to generate arbitrary workloads for OpenStack clouds and to harvest the results,” Ionel said. “So, not only did we create the benchmark, but we created the tooling that allows other people to create their own benchmarks, and generate their own results and publish them back to the community. The reason for this is that transparency and community are very powerful together.”
Ionel said testing and benchmarking will help expand the limits of cloud computing. And while the test results released this week should give enterprises greater confidence that they can build extremely large clouds, he said there’s still tremendous room for growth. “We are not nearly close to the limit. More will be coming. This is just the beginning.”