Working more efficiently with OpenStack

Many years ago I saw someone canoeing using a rather unusual looking paddle whose shaft was bent at an odd angle.  What was that all about?  Why would anyone want a paddle bent at a strange angle?  Aren’t all paddles supposed to be more or less the same, with slightly different shaped blades or made of different kinds of wood? 

As it turns out, a bent shaft paddle gives the paddler more power in the downward draw of the stroke; when pulling the hardest, this bent blade moves a larger volume of water than a standard blade.  Essentially, someone figured out how to make you go faster with the same effort.

So what does this have to do with OpenStack?
We all love standards based, open solutions. They help us build better products, deploy solutions more quickly, and more easily interoperate with products from other companies. On the other hand, pure ‘vanilla’ open source solutions often don’t deliver the performance or the reliability we need in a commercial product. They need some extra work to turn the raw material into the polished, engineered product that’s required.

The Wind River Titanium Cloud virtualisation platform is a good example of such a product. Titanium Cloud is built in part using a number of open source components (Linux, KVM/QEMU, DPDK) and on one of them here lies the focus.

OpenStack
Predictability and trust is very important to companies deploying cloud based solutions for critical infrastructure (e.g. telecommunications and industrial applications).  If an emergency 911 call is made, it must go through. If a valve in a factory plant stays open one second longer than it is supposed to, products can be contaminated or people injured.

As a company with more than 30 years’ experience working with embedded technologies, Wind River understands how to develop solutions with real-time performance characteristics and deterministic execution profiles.

In the OpenStack space alone, Wind River continues to be an active leader, investing thousands of hours of work addressing gaps and bringing new performance features to the OpenStack Nova project.

Wind River’s Titanium Cloud makes extensive use of Intel’s Enhanced Platform Awareness capabilities to help achieve, and exceed, customers’ performance expectations. These Intel EPA features ensure that virtual machines launched on Titanium Cloud compute nodes are optimally placed for the highest levels of performance. 

Placement goes beyond simple NUMA zone awareness to scheduling based on specific CPU attributes (e.g. AES-NI ), NIC locality (same NUMA zone as a specific PCI bus), or the location of peer VMs (collocate or avoid collocation) to name a few. 

Titanium Cloud is even able to isolate VMs from other VMs to avoid the ‘noisy neighbour’ problem where one VM’s use of shared resources could otherwise threaten to impact the performance of other VMs.

Other features available in Titanium Cloud such a memory page sizes, shared virtual CPU and hyper threading policies, even the amount of memory in virtual NUMA zones can be leveraged to gain valuable performance and predictability advantages.

That Wind River would make extensive us of EPA to maximise performance and predictability should come as no surprise, as it is an Intel company.

For more details into the advanced EPA use, refer to the following Wind River whitepaper, ‘Wind River Titanium Cloud: An OpenStack NFVI Product Portfolio with Enhanced Platform Awareness’.

Not all paddles perform the same way, sometimes a few modifications results in large performance payoffs.  In the same way, Wind River’s Titanium Cloud products build on the same OpenStack family, but add important features to make them better suited for use in production, critical infrastructure applications.

Courtesy of Wind River.