by Kishore Jethanandani

Heterogeneous applications and multiple clouds are characteristic of global webscale networks. Traffic flows in such interdependent networks snowball unexpectedly; spikes in application use is endemic which degrades their performance. At its worst, the failure of an application has a domino effect on the network and a catastrophic collapse ensues.

Optimization of web-scale networks irons out their many wrinkles, automates operations, and speeds up responses with predictive algorithms to preempt network outages by deploying resources fast enough to keep pace with anticipated traffic.

Emergence of web-scale networks 

The Twitter Inc. engineering team revealed the details of its redesign for web-scale operations that began following the 2010 World Cup when spikes in traffic disabled its network repeatedly for short periods of time. By August of 2013, Twitter’s infrastructure was robust enough that nothing untoward happened when the traffic surged 20 times over the normal rate during a similar Castle in the Sky event in Japan.

Twitter shifted to a virtualized and microservice-based architecture to gain flexibility. Tweets were assigned an identity so they could be stored on any storage device in a distributed network. Further improvements were made after 2014 to provide route options, distribute resources to the edge and to enable granular optimization with policy management. Similar approaches have been adopted by companies such as Google (Nasdaq: GOOG), Microsoft Corp. (Nasdaq: MSFT) and Facebook .

Investments in bandwidth alone are not enough to cope with traffic flows as they are increasing exponentially due to the growth of the Internet of Things, speech, image, and video data. Web-scale networks streamline processes to avoid local choke points and to increase the overall availability of the network with optimization.

Software-driven performance improvement 

Virtualization and microservices — along with managed services platforms — play a critical role in optimizing the network. Microservices are tools to wring out the inefficiencies by aligning processes with data flows to reduce latencies and increase availability in web-scale networks.

“Microservices are focused on building small services that provide a single piece of functionality,” said Eric Peffer, cloud consulting practice lead, World Wide Technology. “You string several of these microservices together for more advanced functionality. Platforms such as Kubernetes, Pivotal Cloud Foundry, Docker Swarm, Service Fabric and AWS Elastic Beanstalk provide the management and tooling to control the elaborate coordination of the strings of microservices. The data flows are speeded up by abstracting functionality for a series of processes that are aligned to data flows from their source to the destination.”

Software-defined networks also have flexibility in choosing the means to move traffic so that local choke points do not necessarily slow down movement. The operations needed to make or change choices can be executed automatically at the application level.

“There are services available for moving data from one application to another, such as caching, data grid services and message queuing, allowing you to adapt to changes and maintain a consistent flow of data,” Peffer said.

Intelligent network operations 

Web-scale networks interconnect multiple clouds as global enterprises extend the reach of their applications to branches and partners around the world without sacrificing performance. Enterprises want to make their applications and microservices portable across several clouds and interweave them on-demand. Enterprises with a hybrid cloud strategy rose to 58% in 2017 up from 55% in 2016.

VMware Inc. (NYSE: VMW) has built a native cloud on top of Amazon Web Services Inc. for geographical reach, availability, and flexibility. VMware’s multi-cloud management stack, along with its Cloud Foundation and NSX platforms, enables portability across clouds. The bedrock of the management platform is a policy management tool for micro-segmentation of the cloud.

Another survey found that the two most important motivations for multi-cloud strategy was more efficient workloads (73%) and more agility (69%.) Currently, one third of the enterprises want to support multiple clouds for synchronizing applications across them or for workload and data migration. For the future, 42% want most of the resources to be used for management and orchestration across multiple clouds. The intelligence and the management tools are advancing to cope with the increased complexities.

The policy management software plays a supplementary role to the traditional OSS/BSS systems.

“The policies define the parameters for security, configuration, the footprint of the application, edge or core, mini-datacenter, traditional data centers, resource use, networking performance metrics and more,” said VMware’s Gabriele Di Piazza, vice president of products and solutions at Telco NFV Group.

“OSS/BSS systems have been undergoing a significant transformation with IP-based services, which also involved data collection of application and network performance to calculate KPIs,” Di Piazza said. “Machine intelligence does dynamic analysis of data to understand the key determinants of performance to predict network behavior and performance. This is needed to reduce the mean time to repair, take proactive action to prevent failures, or to scale capacity before it falls short. Our acquisitions of Wayfront and Arcane are our investments in real-time gathering of data and predictive algorithms.”

The performance needs to be fortified at every level to maintain a consistency on global networks. Web-scale networks such as Google and AWS have tools to auto-scale in response to surges in traffic; they can spin out new instances when the traffic surges.

“Performance behavior in verticals like e-commerce have their unique characteristics which we identify with our telemetry data,” said Anand Hariharan, vice president of products, Webscale Networks Inc. “The traffic can surge by as much as a hundred times following events like celebrities posting a picture on Instagram with their products. We have written an algorithm to forecast traffic surges specifically for e-commerce to deploy more instances to keep pace with demand growth.”

Web-scale networks are buffeted by a myriad of factors at the local and global level in addition to the demands of new applications and sources of data. They have more choices of routes and network configurations for design and automation tools to adapt in the moment. Machine learning will continue to evolve as more data is gathered and innovative ways are found to direct and optimize traffic.

A version of this article was published by Light Reading’s Telco Transformation

---

---