Introduced this year, Google Network Service Tiers let cloud users trade off between price and performance, but...
it's important to understand Google's network architecture, and your own cloud networking requirements, before you choose a tier.
Google operates one of the largest private, global WANs in the world and uses a customized, OpenFlow-based software-defined network to route traffic between its data centers. Google operates two distinct WANs: one that faces users and network peers, such as internet service providers (ISPs) and carriers, and an internal network -- called B4 -- that connects its data centers.
Google divides and prioritizes traffic into three classes based on volume, latency sensitivity and overall application criticality. It also uses a software control plane to optimize Google Cloud Platform (GCP) traffic flow. Google's internal WAN is designed for massive volume and minimum latency, so Google attempts to maximize use of that WAN over public internet circuits for its commercial applications, like YouTube or Google Drive.
With its Google Network Service Tiers, the company makes these advantages more explicit to users.
An intro to Google Network Service Tiers
GCP's original method of routing WAN traffic -- now called the Premium Tier -- uses its software-defined WAN to onboard inbound user traffic to the Google private WAN at the closest point of presence (POP). Once on Google's private WAN, the service routes GCP traffic to the region and data center that contains your application. Outbound traffic works identically in reverse; traffic leaves the GCP data center that runs the application, routes over Google's private WAN to the POP that's nearest to the application user, and exits onto the public internet for final delivery.
The Standard Tier, which works like traditional cloud provider networks, routes user traffic over public ISPs and carrier networks to the closest POP within the region that hosts a user's GCP services. This reserves Google's private WAN for connections only within the region. Again, the process works the same in reverse.
One of the significant differences between the two Google Network Service Tiers is how they deliver server load balancing. The Premium Tier provides globally distributed load balancing and the Google Cloud Content Delivery Network (CDN). In this model, load balancers advertise a single IP around the world and direct traffic to the Google POP closest to the user, routing over the private WAN to the data center that hosts GCP instances.
Aside from the convenience and simplicity of having a single address for a website, global load balancing enables organizations to automatically use systems in other regions as backup capacity. In contrast, the Standard Tier provides load balancing only within a region and doesn't include the Google Cloud CDN service.
Premium Tier users also get higher availability. Its private backbone uses at least three independent paths (N+2 redundancy) between any two Google locations, which means that traffic is unaffected if one or even two fiber lines are lost between data centers.
According to Google, unlike the Standard Tier, the Premium Tier will have a service-level agreement that guarantees global performance.
Performance and pricing
Google quantifies its performance claims with real-time measurements that show the Premium Tier provides about 67% faster median throughput (5 Mbps vs. 3 Mbps), and a 13% reduction in latency (80 ms vs. 92 ms), compared to the Standard Tier. Those numbers are likely to change since the tiers are still in alpha, and Google only performed these measurements in a single region.
Pricing varies by region and volume. However, within North America and Europe, the Premium Tier is about 23% more expensive per gigabyte delivered than the Standard Tier for volumes up to 500 TB per month, with custom prices available for high-volume users. The Premium Tier pricing represents up to a 25% cut from GCP's existing network pricing, so the Standard Tier provides a significant discount from current rates.
To determine which of the Google Network Service Tiers is right for you are, ask these three questions:
- Does my service require the highest performance and availability?
- Is my back-end data traffic sensitive enough that it's best to keep it off public internet circuits?
- Will my service run in multiple regions around the world and possibly need global load balancing and a CDN?
If none of these apply, the Standard Tier is the better option, because it saves money. Otherwise, it's better to pay up for the Premium Tier.
Choose the right network services for hybrid cloud
Explore Azure network options and services
Is software-defined networking in hybrid cloud the answer to cloud networking problems?