Microsoft Azure feature targets application latency woes

While many companies spread their cloud deployments across multiple regions for increased reliability and recovery, keeping virtual machines close together in the same data center can reduce application latency.

That is the point of a just-released Microsoft Azure feature called proximity placement groups. Network latency is a key component of application performance, and given the limitations of the speed of light, the physical location of virtual machines can contribute to application latency.

Azure customers can reduce application latency by putting VMs in a single region, and go further by placing them in a single availability zone, Microsoft said in a blog post.

"However, as the Azure footprint grows, a single availability zone may span multiple physical data centers resulting in network latency that can impact your overall application performance," Microsoft said in the blog post.

Proximity placement groups aim to mitigate this issue. Customers can assign VMs for an application's various tiers to a proximity group, which ensures they all end up in the same data center. This works even for VMs that are a mix of sizes and types, according to Microsoft.

Customers can adopt proximity placement groups via Azure Resource Manager templates. Customers should use accelerated networking and optimize their VMs to reduce application latency most effectively, Microsoft said.

It is also a bit tricky to apply proximity placement groups to elastic deployments that scale out on demand. Customers can avoid this if they ask for all the VMs to be placed in a proximity group at the same time, according to Microsoft.

Gary Chen

The feature is now in preview at no cost, and available in all Azure regions apart from Japan East, India Central and Australia East. Microsoft is not first to market with the capability, as AWS already offered Placement Groups, which has a similar intent.

Microsoft wants to help customers with applications that have a lot of east-west traffic, a term that refers to data that flows within a data center, rather than to and from multiple data centers, said Gary Chen, an analyst with IDC.

If you have a high-performance workload that has a lot of that east-west traffic, this would be a big deal to you.

"In a data center, you're used to being able to do stuff like this with VMware," Chen said. "If you have a high-performance workload that has a lot of that east-west traffic, this would be a big deal to you."

One example of this is an SAP ERP application, which has a three-tier infrastructure that spans the presentation layer, application server and database. ERP applications are computationally intense, and require constant interplay between the tiers, particularly the app server and database.

Indeed, SAP participated in the feature's early preview program. The company managed to reduce application latency between SAP components to less than 0.3 milliseconds, according to a statement included in the Microsoft blog.