Google unveils Andromeda 2.0
Google announced it has made some updates to software-defined networking (SDN) platform Andromeda, which the company said should reduce network latency between Compute Engine virtual machines (VMs) by 40% and almost a factor of eight since launched in 2014.
Andromeda is a SDN-based substrate for the company’s virtualization efforts. It serves as an orchestration point for provisioning, configuring and managing VMs, and in-network packet processing. Google’s Cloud firewalls, routing and forwarding rules all leverage internal Andromeda application program interfaces (APIs) and infrastructure, according to the company.
Poor network latency delays how quickly data is transferred from one location to the next, creating problems for companies with customers that expect a positive end-user experience. The latency improvements attached to Andromeda 2.1 are the result of a kind of hypervisor bypass that builds upon virtio, an abstraction layer over devices in a paravirtualized hypervisor included in the mainline Linux kernel.
A paravirtualization hypervisor consists of a guest operating system that is aware it is running on a hypervisor and includes a code for making guest-to-hypervisor transitions more efficient. By contrast, a full virtualization system scheme consists of a guest operating system running atop of a hypervisor sitting atop bare metal. Andromeda 2.1 allows the Compute Engine guest VMs and the Andromeda software to communicate directly with shared memory, circumnavigating the hypervisor in return.
“This kind of network performance is especially important as more applications move into the cloud and are accessed via web browsers,” wrote Jake Adriaens, staff software engineer at Google, in a company blogpost. “While the headline metric is often bandwidth, network latency is frequently the more important determiner of application performance.” He added low latency is essential for tasks related to financial transactions, ad-tech, video, gaming and retail and workloads.
In addition to updating Andromeda, Google announced this week updates to its Dedicated Interconnect, which provides a way for businesses to extend their corporate network into Google Cloud. Dedicated Interconnect provided 35 interconnect locations when it was first released in September, adding four more locations in Atlanta, Mumbai, Munich and Montreal as part of the update.
John Veizades, Google product manager, noted in another company blog post that Dedicated Interconnect is now generally available, ready for production-grade workloads and covered by a service level agreement. The service has also been updated to support global routing for Cloud Router, a new feature that permits subnets in GCP to be accessed by an on-premise network via the Google network.
