Turbo boost your Compute Engine workloads with new 100 Gbps networking
Software Engineer, Google Madison
Today, we are excited to announce the public preview of 100, 75, and 50 Gbps high-bandwidth network configurations for General Purpose N2 and Compute Optimized C2 Compute Engine VM families. This is the result of continuous efforts to optimize our Andromeda host networking stack, allowing us to offer higher-bandwidth options on existing VM families when using the Google Virtual NIC (gVNIC). These VMs were previously limited to 32 Gbps.
Some of the most demanding workloads on Google Cloud will now be able to take advantage of these high-throughput VMs, such as tightly-coupled high performance computing (HPC), network appliances, financial risk modeling and simulation, and scale-out analytics. Combining high-throughput VMs with high-performance Local SSD will be beneficial for I/O-intensive, flash-optimized databases. For applications that are sensitive to both latency and throughput, you’ll be able to combine compact placement policies on C2 instances with 100 Gbps bandwidth for superior networking performance.
Flexible bandwidth options
We’ve always offered flexible packaging options to give you control over the compute resources that work best for your needs. We’re continuing that trend by offering these higher bandwidth configurations as add-on options for existing, mid- to large-size N2 and C2 machines.
The tables below summarize the new networking options for those VM families:
For N2 VMs that have standard, high-memory, high-cpu, and custom configurations, you can expect these same bandwidth options as long as they meet the above vCPU requirements.
Since these higher bandwidth configurations are optional, add-on features for your VMs, they will show up as incremental charges over and above what you pay for the underlying VM, as separate dedicated network bandwidth SKUs on your Cloud Billing report. We meter VM network bandwidth the same way we do for vCPUs and memory, in dollars per hour. Visit our pricing page for specific pricing in your region.
Here’s an example using the beta gCloud SDK to create a N2 instance with 75 Gbps network bandwidth:
Using the TIER_1 setting on the network-performance-configs flag automatically upgrades your instance with increased network bandwidth.
To use these new features, make sure that you’re using the beta channel, and run gcloud components update to get the latest SDK. See Benchmarking higher bandwidth VM instances for more information on setting up VMs and running network throughput tests.
We strive to make it easy for our customers to take advantage of our differentiated features. Some unique aspects of high-throughput VM networking are:
It’s an add-on feature. Since we’ve optimized networking performance on existing N2 and C2 VMs, you don’t need to reconfigure your workloads for any new instance types to take advantage of high-throughput networking. You also maintain all compatibility with scripts and automation tools.
No additional inventory constraints. These networking capabilities don’t impose additional inventory constraints on your N2 or C2 deployments. In fact, you can upgrade VMs to use increased bandwidth in any zone where you can create an N2 or C2 instance.
Best-in-class throughput. Google Cloud offers the best throughput performance of any cloud provider, and we’re doubling down on that performance with even more bandwidth offerings today.
Leveraging Google Cloud's unique Andromeda architecture
Our new A2 instances with NVIDIA A100 GPUs support 100 Gbps networking today. Building on our work for A2s, we’re now able to offer higher bandwidth configurations on existing C2 and N2 VM families by optimizing Google Cloud’s unique Andromeda network, which we’ve upgraded to support hardware offloads such as zero-copy, TSO, and encryption, all without introducing downtime for your VM. The result is up-to 100 Gbps of encrypted traffic between VMs.
Get started today
High bandwidth configurations for N2 and C2 VMs are available in Preview today (in regions and zones that support these machine types) for all customers using the beta gCloud SDK or our Cloud APIs.
Please note that this feature is in public preview and will be billed at no cost throughout this period. The pricing in our documentation is general availability pricing and will go into effect when the public preview ends.