YouTube, Gmail, Maps, and Search. What do they have in common?
The answer is an underlying infrastructure that is built to scale – and now available for any enterprise, including yours, by Google Cloud.
Too often, cloud infrastructure management is beset by continually shifting trade-offs as IT teams seek to realize the benefits of scale. Google Cloud has simplified the process and boosted the performance of applications — at no extra cost — with its Titanium system, which offloads key tasks to free up valuable IT resources.
This is a big deal. Managing cloud infrastructure can be frustrating because it’s a struggle to optimize both workload performance and costs. Efforts to improve efficiency and optimization in one area could result in bottlenecks or poor performance in other areas.
Traditionally, the CPU has been responsible for managing everything – the workload and all platform tasks that supported it, including hypervisor processes, the virtualization stack and management of storage and networking I/O. So, cloud workloads were in contention for host CPU resources.
With Titanium, many of these tasks are distributed across a system of purpose-built hardware, custom silicon, security microcontrollers, and tiered scale-out offload processors. Titanium enables cloud workloads to use datacenter-wide resources and, with many infrastructure tasks now automatically processed off-host, CPU resources are freed up for improved workload performance. As a result, Google Cloud can provide more compute power to its customers at the same price.
The same infrastructure that powers Google’s scalable apps
Google powers some of the largest and most scalable services in the world including Search, YouTube, Gmail, and Maps. With Titanium, Google Cloud runs on the same infrastructure:
- The Jupiter scalable data center fabric is its networking backbone, seamlessly connecting resources.
- Google’s Borg cluster management system orchestrates millions of jobs across vast clusters of machines.
- A globally distributed, exabyte storage system, Colossus, underpins essential cloud services and provides the foundation for all persistent disk drives attached to virtual machines (VMs).
Titanium works across every new Google Cloud VM instance to provide the same networking, storage, and management capabilities for a diverse set of workloads. That includes AMD’s EPYC-powered C3D VMs.
Datacenter-wide offload processing
The first step in the Titanium system is on-host, where PCIe cards take on networking and storage tasks, while security processing is sent to Titan microcontrollers that ensure systems boot from a known, trusted state using verifiable code.
Off-host, Titanium enables cloud workloads to use data center wide resources, without applications having to do anything. Each host CPU is connected to a series of Titanium Offload Processors throughout our data center that form a second tier of relief for customer applications.
This second tier is made possible thanks to a management layer that schedules resources and processing throughout the data center to free up more performance capacity for the host CPU. Titanium was designed to be updated without impacting customer workloads. It uses a combination of techniques from hit-less updates of hardware and software components to VM live migration.
With on and off host layers working together, Titanium enables Google’s C3D offering with Hyperdisk Extreme to provide 360K IOPs per instance and 200 G/bps of network bandwidth.
Titanium cloud infrastructure results in enhanced performance, greater flexibility, and reduced complexity – enabling IT to achieve the full potential of workloads – providing added scale without having to revise subscriptions.
Learn more here.