One of these is a degree of stickiness due to the technical friction, architectural complexities and costs of moving from platform to another. Organizations that commit a significant share of their IT fleet to a platform like AWS or Azure don’t exactly face Hotel California-style lock-in, but they do risk a form of IT golden handcuffs that makes it far easier to stay, than switch, even when an alternative better meets their changing needs or the service becomes otherwise less competitive.
There are several barriers to switching between cloud providers, however the two most significant are:
- Dependence on proprietary or idiosyncratic services, such as managed container services as I recently discussed, incompatible event-driven serverless functions, unique AI or data analytics services or proprietary `cloud APIs.
- Data gravity, an apt metaphor coined almost a decade ago, that covers the time, cost and effort of moving larger and larger cloud-hosted data sets.
Although the gravitational metaphor nicely summarizes the non-linearly increasing effort needed to move more massive data repositories, it glosses over the very real financial (not just personnel) costs of doing so. Sticking with the planetary imagery, just as it takes far more rocket fuel to escape the earth’s gravitational pull than the moon’s, the more data you move from a cloud service, the larger your network bill due to the network egress charges each imposes.
For example, copying 10 TB from AWS S3 in US East to another service via the Internet (presumably using a VPN), the equivalent of just over 300 GB per day, costs $900. While organizations using Direct Connect can cut the bill to $200 (at least in the U.S. but higher elsewhere), they must also factor in monthly port connection fees from both AWS and a remote cross-connect service provider like Equinix or AT&T.
In sum, today’s standard cloud service billing model resembles those moving sidewalks in Las Vegas: you get a free ride on the way in, but they'll make you sweat on the way out. While there are legitimate reasons for charging something, since the cloud providers do incur expenses for Internet circuits, associated network equipment and cabling, they aren't close to being as high as current egress fees imply. Indeed, network egress is a high margin cash cow akin to the outrageous implied cost of storage on a smartphone where Apple, Google and others charge $150 for 128-192 GB of memory that costs less than $50.
An alliance for data freedom
Amidst this environment of onerous data charges, Cloudflare emerged as an opposing force driving to reduce network costs and improve performance and reliability through its Internet transport, content delivery and other network services. Last week, the company initiated a significant program to improve cloud data freedom via the Bandwidth Alliance, a group of cloud and networking companies committed to reducing data transfer fees.
Companies in the Alliance, which aside from Cloudflare includes Microsoft Azure, IBM Cloud, DigitalOcean and seven others, agreed to reduce or eliminate bandwidth transfer charges for their mutual customers. Although the technical and accounting details are yet to be completed, Cloudflare estimates that its existing customers could collectively save about $50 million per year under the agreement.
Notably absent from the Alliance are two of the three cloud giants: AWS and Google Cloud. Both have been customarily silent on the announcement, but Google Cloud is missing for a legitimate reason: for the past three years the company has worked with Cloudflare on a CDN interconnect program that Cloudflare says cuts transfer fees between their mutual customers by up to 75 percent. The result is that Google Cloud customers already get many of the promised Bandwidth Alliance benefits. Indeed, a more likely reason Google opted out is the principal technical component of the Alliance: a new Cloudflare-developed routing protocol designed to optimize traffic flows between members and automate monitoring and accounting to ensure that customers get credit for the discounted rates.
Building on Cloudflare's network technology
Cloudflare discusses some of the technical factors that enable the Alliance, notably its existing global network with more than 150 POPs and peering relationships with major Internet exchanges, network providers and private network operators that provide multiple paths between cloud service providers and customers. Controlling Cloudflare's network mesh is a custom-build routing engine called Argo that decides how best to carry and switch traffic across the network while optimizing for latency, throughput, jitter and now, under the Alliance, cost.
Cloudflare built Argo to remedy deficiencies in BGP, the standard routing protocol used by every organization connected to the Internet. BGP attempts to choose an optimal path between two destinations, which typically means that with the fewest number of network hops. Unfortunately, as the Cloudflare post points out, the protocol has no means of knowing the cost or performance of a particular link. Thus, BGP can’t prioritize links for a company like Cloudflare, with access to thousands of different networks, some using private, fast and cheap fiber, others on congested public circuits.
Argo improves on BGP by digesting Cloudflare’s network topology and performance metrics to find the path with the lowest latency and fewest hops on public circuits. For Bandwidth Alliance members, Argo can choose paths that stay within Cloudflare’s network to the greatest extent possible, which both improves performance and reduces cost.
Google, which operates perhaps the world’s largest private network with hundreds of POPs and enough fiber to circle the globe a few times, has long had technology similar and likely superior to Argo. Indeed, Google’s Premium Network Tier maximizes traffic flow over its private network with most in- or outbound traffic reaching its destination with a single hop from the Google network to an end user’s ISP. Given the existing relationship between Google and Cloudflare, if a Google customer’s final destination is a Bandwidth Alliance member, the last hop would be to Cloudflare and therefore benefit from the existing CDN discounting scheme. Thus, it appears Google Cloud can deliver most of the benefits of Cloud Alliance membership without being party to Cloudflare’s Argo routing system.
The Bandwidth Alliance promises to reduce a significant deterrent to implementing a multi-cloud sourcing strategy, namely the costs with shuttling data between disparate services. Cloud egress fees far more than the actual costs of providing the service are like a tariff on data. Eliminating or reducing them promises similar benefits to tariff-reducing multinational trade agreements, namely it promises to increase overall usage, increase the diversity of services, improve efficiency through specialization and scale and reduce the cost to consumers via increased competition.
Indeed, as the mega cloud providers continue to get larger, adding more and more services, its easy to forget Ricardo’s theory of comparative advantage, i.e. if there are few barriers to using multiple suppliers for one’s needs, it encourages firms to maximize the output of goods or services were they have a competitive advantage and outsource those where they do not.
Applied to the multi-cloud scenario, frictionless data movement between providers should facilitate the emergence and growth of firms specializing in particular services where they have a comparative technical or business advantage (such as a deep understanding of an industry or problem domain) since there will little barrier to customers moving information stored on a major provider such as Azure or Google, to a third-party for additional processing, analysis or visualization. Of course, such a scenario is counter to Amazon’s horizontally-integrated model, both in retail and cloud services, which likely explains its absence, at least for now, from Cloudflare’s consortium.
The Bandwidth Alliance is an essential step towards multi-cloud integration within an organization and could unleash a wave of specialized secondary cloud services that supplement and improve upon the capabilities of the major cloud providers. I hope the effort gains membership and mindshare among both customers and as-a-service providers.