The most important qualities to look for in an IaaS platform include cloud computing performance benchmark results, a flexible pricing plan and user-friendliness. A few weeks ago I posted a discussion about user-friendly data center configuration and management via our unique Data Center Designer. This time I want to share some insight about ProfitBricks’ high performance cloud, with interviews from three members of our engineering team to illustrate in detail how we achieve the fastest cloud computing performance available.
“Optimum cloud computing performance”
Our combined use of InfiniBand (the network interconnect technology that connects our servers and servers to storage at 80 Gbps) and our virtualization layer technology KVM results in a fast, reliable and scalable cloud for which we’re proud to receive external recognition.
When we won Frost & Sullivan’s Best Practices Award for our innovative cloud computing solution, their accompanying report praised ProfitBricks’ platform for delivering “optimum cloud computing performance along with flexibility in resource allocation and suitable billing procedures,” thus enabling our customers to “enjoy the limitless nature of cloud computing.”
InfiniBand: high speed network connections
“InfiniBand connections are usually reserved for “high performance computing (HPC),” says Sebastian Parschauer, senior Linux kernel developer at ProfitBricks. “We are able to transfer at 80 Gbps using two ports simultaneously.” ProfitBricks customers get free InfiniBand conenctions between their server and between their servers and storage. They can create clusters and transfer data at far higher speeds than any other cloud.
According to the standardized UnixBench benchmarking test, ProfitBricks is at least 2.3 times faster than Amazon AWS, and using the Iperf tool, ProfitBricks is at least 7.3 times faster. Alongside these widely used open source tools, we recommend also carrying out a workload-specific benchmarking test to measure our cloud computing performance for your individual use case. Though they offer a 10 Gbps option, Amazon’s default is 1 Gbps and sometimes, due to throttling on their side, their speeds drop as low as 500 Mbps. And because they use older Ethernet technology, they have higher latency and more inconsistencies than the ProfitBricks InfiniBand powered cloud.
ProfitBricks’ head of network operations, Torsten-Sven Urbas says: “We are the fastest cloud provider, not just because of the high throughput but also because of our low latency. While Ethernet devices typically have a latency of 550 nanoseconds (ns), with exceptional Ethernet switches providing 350 ns, the Mellanox InfiniBand Switch that ProfitBricks uses has a latency of less than 100 ns.”
What other edge do we have over cloud providers who use regular Ethernet? “The end-to-end latency is also low,” says Sebastian, “which means we don’t spend much time waiting for signalling responses. The secret behind this is remote direct memory access (RDMA) implemented into hardware. In contrast to regular Ethernet, the operating system isn’t involved in the data transfer. This decreases CPU utilization by 8 times. So InfiniBand with redundant fabrics means we can offer quick and reliable networking and storage.”
And the benefits of InfiniBand don’t stop there. Sebastian lists several other benefits, one of which is the fact that “InfiniBand self-manages, so we can hot-add more nodes to our fabrics and they become usable immediately if customer demand increases. In case of an issue, alternative paths are chosen automatically and diagnostic commands can be run on any node for the whole fabric, meaning no interruptions or downtime.”
KVM: flexible performance with zero downtime
Our InfiniBand connection works together with KVM (an open source kernel-based virtual machine), enabling our customers to vertically scale their server instance sizes, from 1 to 62 CPU cores and 1 to 240 GB RAM. “Flexibility is one of ProfitBricks’ main selling points. Offering dedicated CPU cores and dedicated RAM means we are always ready to deploy more processing power and speed on demand,” says Anshul Makkar, paravirtualization engineer at ProfitBricks. “With our Live Vertical Scaling feature, customers can upscale their CPU cores and RAM at runtime with no reboots, it’s great for example when they are expecting more or less visitors or users than usual.”
ProfitBricks also includes standard double redundant RAID 10 block storage; we don’t offer temporary or ‘ephemeral’ storage. ProfitBricks block storage creates four copies of each block of data in two separate physical locations meaning customers’ data is never shared between server instances. Because of this, ProfitBricks can enable live migration of the cloud and servers while other IaaS providers sometimes force downtime and cloud reboots on their customers.
“We dedicate storage space to replicate customers’ data which protects customers against losing valuable data. If a running virtual machine needs to be live migrated to a new server, it can be transported from one physical host to another at such speed that our customer won’t even notice. They’re up and running all the time,” Anshul says.
The leader in cloud computing performance and pricing
Using InfiniBand as our high speed network connection and KVM as our virtualization layer we can offer flexible, secure and reliable cloud data centers. These features have enabled ProfitBricks to be the price/performance leader in IaaS cloud computing since we launched in 2012. As a result, our customers benefit from optimum cloud computing performance, zero maintenance downtime and our cost-efficient minute-based billing system. Features like this helped Frost & Sullivan to pick ProfitBricks as their innovation leader in cloud computing for 2014.
What kind of cloud computing performance have you experienced so far? Have you run any performance benchmarks of your workloads? Sign up for a free trial with ProfitBricks and test your workloads today, and please contact us if you need any help configuring your cloud for testing.