Cloud computing has fundamentally altered the tech landscape, enabling an unprecedented level of connectivity and processing power. With processing power or compute resources increasingly becoming a commodity service, premium connectivity products have taken rise to stand out from the flock. This is with good reason, because as the digital highway propels our increasingly virtual world, these services promise low latency, no jitter or packet loss and connection that is as secure as possible. Businesses around the globe leverage these services to ensure seamless and resilient user experiences, be it for video conferencing, online multiplayer games, streaming services or virtually any other web-based application.
As enticing as premium network products are, they come with an equally daunting obstacle: the additional cost on top of the already “premium” egress cost for standard network products. ‘Egress cost’ is a term used to describe the charges associated with data leaving a public cloud provider’s network, which needs to be a key consideration for businesses running high-traffic volume applications, as this will come with significant costs.
Most public clouds have had no issues with charging high costs for egress. At the end of the day, their purpose is to keep as much data within their own ecosystems since they offer a vast array of services hence, “punishing” any data leaving their ecosystem. And yes, the invoice at the end of the month can hurt.
Despite high egress costs and the additional cost that premium network products put on top, there are many organizations that prefer to leverage premium network products. Why? There are plenty of reasons, but it boils down to the need for consistent, high-speed and reliable connectivity that scales along with your applications. The cost of a network outage or slow connection far outweighs the financial expense of egress costs and premium network products as it may impact user experience, damage your brand reputation and end up losing potential business if your application is not reachable.
With that, let’s dive into two public clouds and the premium network products they offer on top of their regular connectivity and egress costs: AWS Global Accelerator and Google Cloud Platform’s (GCP) Premium Network.
AWS Global Accelerator is a network service that simplifies traffic management and improves performance by up to 60%i. It provides users paying a premium — on top of the regular AWS egress costs — a way to bypass internet congestion, optimize the network path and boost data transfer speed.
The service directs traffic to an optimal endpoint based on the application’s health and network conditions. This ensures improved availability and performance, reducing the latency of applications by the stated 60% and improving resistance to DDoS attacks. On top of that, it has additional features such as anycast capabilities, TCP edge termination and custom routing accelerators.
For real-time applications like multiplayer games, previously mentioned software features in the AWS Global Accelerator are unnecessary as the core importance for multiplayer games for instance, is to ensure that the traffic between the gamer and the game server stays as local as possible, and if not, it travels over a private backbone to avoid public internet congestion (which may result in a sub-optimal player experience). This means that the only aspect of the AWS Global Accelerator relevant for game servers would be the announcement of IP addresses from multiple AWS edge locations at the same time to ensure global performance-based routing over their backbone.
GCP also offers a solution to enhance the GCP Standard Network, which is dubbed the “GCP Premium Network”. This service is designed to deliver robust performance, providing low-latency, high-throughput and reliable connections at a premium on top of the regular egress cost.
Unlike GCP’s standard network, which uses the public internet over long distances, its Premium Network leverages Google’s private global fiber and its edge-caching services. This effectively allows traffic to enter Google’s network as close to the client as possible, traveling over Google’s dedicated infrastructure and exiting as close to the destination as possible, resulting, in theory, in faster and more secure data transfers.
According to their website, in GCP’s standard network, their routing is cost optimized rather than performance optimized. Referring to the images below from GCP’s website, the differences between the two form a stark contrast. This is especially the case for real-time applications such as video conferencing software, which if hosted on GCP would have a sub-optimal user experience should they not upgrade and pay extra for the premium network.
There are a few disadvantages to premium network products of public cloud providers, but aside from these one may beg the question as to why they do not enable ‘cold potato routing’ or focus on performance by default. That said, the two most present downsides would be:
One thing is certain — network performance issues are a real problem, especially for real-time applications. This allows public cloud providers to monetize their premium network products. The question is, how premium are these premium network products, or are their standard network products simply underperforming for real-time applications? This would therefore force their customers to pay a premium for things like cold potato routing. Public cloud providers such as GCP and AWS have gotten to a size where their standard operating mode must be hot potato routing, as the bulk of their traffic isn’t reliant on the best network performance and if they would change this practice, their egress costs would rise even further.
Like public cloud providers, i3D.net also has built its own global network including a private backbone, connected to 100+ Internet Exchange Points and peered heavily with local and global networks. However, at i3D.net we do not charge additional cost for so-called “premium network products”. Our standard operating mode is to prioritize performance rather than cost optimization.
The interesting thing there is that even though we prioritize performance on our network, our egress cost is a fraction of those of the public cloud providers since our business model isn’t to capture and then keep you in the ecosystem, but rather ensure your end-user experience is the best in class while we take care of the underlying IT Infrastructure to support it.
That means that i3D.net does not discriminate between local and remote routes by leveraging BGP and IGP, ensuring data packets on our network always take the best path. We are able to do this by being the 2nd most adjacent network, with 9000+ possible entry points globally. And when traffic does enter the i3D.net network as close to the end-user as possible, it is prioritized for the lowest latency achievable.
AWS standard network
AWS Global Accelerator
GCP Standard Tier Network
GCP Premium Tier Network
Standard routing method
No discrimination between local and remote routes
Routing optimized for
BGP peers observed*
441 (IPv4) 341 (IPv6)
499 (IPv4) 387 (IPv6)
8829 (IPv4) 5747 (IPv6)
AWS Shield Standard (no extra cost)
AWS Shield Standard (no extra cost) & AWS Shield Advanced (optional paid)
Standard Cloud Armor (pay as you go) & Managed protection plus (paid subscription)
Standard Cloud Armor (pay as you go) & Managed protection plus (paid subscription)
GLAD Basic (no extra cost) GLAD Advanced (optional paid) GLAD Premium (optional paid)
In conclusion, premium network products have value for certain applications if they’re willing to accept the additional costs on top of the egress cost. Ultimately, choosing the right service comes down to understanding your unique needs and weighing the costs against the benefits.
Prior to selecting your compute provider — as you will have to use their network services — It will pay off to do your due diligence on their network and the associated costs. For some applications, specifically real-time applications, you’ll need a network as heavily peered as possible, optimized for performance, with cold-potato routing, its own backbone and with consistent IP address announcements.
Prior to selecting your compute provider — as you will have to use their network services — it will pay off to do your due diligence on their network and the associated costs. For some applications, specifically real-time applications, you’ll need a network as heavily peered as possible, optimized for performance, with cold-potato routing, its own backbone and with consistent IP address announcements.