Google has released a new version of its Andromeda software defined network (SDN) stack that among other things significantly speeds up network performance for customers of the company’s cloud computing services.
Andromeda 2.1 reduces latency between virtual machines on Google’s cloud platform by 40 percent compared to the previous Andromeda 2.0 stack and by a factor of eight over the original version introduced in 2014, according to the company.
“This kind of network performance is especially important as more applications move into the cloud and are accessed via web browsers,” Google staff software engineer Jake Adriaens announced on the company’s cloud platform blog Nov. 2.
While many organizations typically tend to focus on bandwidth as a key metric for application performance, network latency has more of an impact, according to Adriaens.
Financial applications and those in areas such as retail, advertising, gaming and high-performance computing can all benefit from lower latencies in the cloud, he noted. In memory databases and HTTP-based microservices should similarly substantially benefit from the latency reductions.
Google’s Andromeda is an SDN technology that is essential to the company’s cloud network virtualization capabilities. The company has used the SDN technology for years internally to manage the provisioning and configuring of virtual networks.
The technology enables software-defined control over the various components of Google’s cloud network stack including switches, routers, network interface cards, virtual machines, operating systems and hypervisors.
Google has described Andromeda as enabling several critical cloud management capabilities including better resilience against distributed denial of service attacks, load balancing and access control lists. The Andromeda SDN also enables easier rollout of new network cloud services and security patching, the company has said.
The latency improvements in the latest version of the technology come mainly from a hypervisor bypass feature that enables a virtual machine on Google’s cloud to communicate directly with the Andromeda software switch, Adriaens wrote. Previously, this communication between a VM and the software switch happened via the hypervisor, which was a slower process.
The hypervisor bypass feature substantially reduces VM to VM network latencies and speeds up round trip times for packets travelling between VMs. “This reduction in network round-trip times translates into real-world performance boosts for latency sensitive applications,” Adriaens said.
Andromeda is designed to extract maximum raw performance from the underlying network. As a result, most organizations in fact, should see, intra-zone or VM to VM latency improvements regardless of the applications they have in the cloud, according to Adriaens.
Andromeda represents just one component of a broader Google SDN strategy. Earlier this year Google revealed an SDN technology called Espresso, which it uses for directing traffic at the network edge.
More than 20 percent of Google’s traffic to the Internet is already being handled by Espresso. Other core SDNs running on Google’s cloud include B4, which powers the company’s datacenter Wide Area Network and Jupiter for data center networking. Together, Google has described Andromeda, Jupiter, B4 and Espresso as the pillars of the company’s SDN stack at the company.