CNET's Gordon Haff wrote a great piece on the shortcomings of the exchange model for cloud computing. His prognosis is right there in his title: "Why cloud exchanges won't work." I've done some thinking and writing on the topic, and it's easy to see Haff's point: Interoperability, security, and inertia threaten to derail this new concept before it starts.
Shortcomings of the Exchange Model
Haff's concept is centered on the following three simple qualifiers for cloud exchanges:
- Any platform involved in an exchange must be compatible, allowing a workload to seamlessly move between interoperable systems. This is both critical and absent with many of the cloud computing services available today. Most are incompatible on a basic level, using different hypervisors for example. No cloud exchange can seamlessly move an EC2 Xen instance to a Terremark VMware environment, although Rightscale is working on clever translation systems.
- Not all service providers are equal when it comes to security and compliance, either. I've often bemoaned the fact that so many service providers are not enterprise-ready, and this will be even more of an issue with an intermediary deciding where to run a particular job. How can a buyer be sure his workload will be safe?
- Haff also points out that a compelling service must be cost-effective, and certain elements stand in the way of this. He questions the value economies of scale will bring to very-large service providers. He also wisely points out that additional costs to move supporting stored data could derail the return on investment
These are not intractable problems, but they are real concerns. The issue of portability is especially thorny for cloud computing, as vendors focus more on basic functionality and innovative features than compatibility. Yet one can envision a future in which even these issues are resolved: Haff worries about service providers moving "up the stack", but this is exactly where compatibility is likely to emerge. It is easier for me to imagine a number of interoperable Java or .NET platforms than truly compatible Xen environments!
It also seems that the current efforts to define cloud service description and provisioning APIs addresses many of these concerns. If a standard API could specify compatibility, the network environment, and security requirements, an exchange could offer a wide variety of service providers with these capabilities. I imagine an airline model, where not every airport is served by every airline, but there is enough competition even at the fringes to keep the resellers viable.
I am also much less concerned about return on investment than Haff. I have seen amazing economies of scale achieved at Nirvanix (where I am Director of Consulting) and can imagine these in many areas. Cloud service providers are driving cost out of many areas by standardizing and centralizing management operation and provisioning as well as hardware and environmental costs. As cloud providers set up shop in super-efficient data centers and train highly focused management staff, they are likely to surpass the economies of even the largest abd best-run end user environment.
What About Storage?
As I wrote in December, spot pricing for cloud storage is much less attractive due to the sheer inertia of data. But an exchange model might actually be attractive even after this is taken into account. The major public cloud storage providers have already moved up from the infrastructure (bytes and blocks) to the platform (object) level, though the "ammo provider" private cloud market remains focused on the former. And many efforts are already underway to create basic standard interfaces for both provisioning and access across these major vendors.
Let's take Haff's concerns in order:
- Interoperability of cloud storage is likely to come well before compute thanks to the more constrained workload involved. While a compute platform could be asked to perform almost any task, storage in general is focused on a simple usage model sometimes called CRUD: Data is created, read, updated, and deleted. This has already led to a proliferation of pre-standard or de-facto standard generic interfaces to multiple cloud storage services. SNIA's rapid work on a standard cloud data management interface shows that interoperable public cloud storage isn't that far off.
- Security and compliance is much less standardized among private and public cloud storage providers. I believe that all enterprise-focused public cloud storage providers should focus their efforts on offering solid, reliable, and secure systems, but this has clearly not been the case universally. And although efforts like the SNIA CDMI would standardize provisioning of services, we will need a much more robust vocabulary to specify the level of security and compliance required for a specific application. But not every application is right for cloud storage anyway. Surely a minimum standard for security can be agreed upon by multiple providers, allowing at least some bulky applications to use cloud storage without worry. We will eventually develop a more complete mechanism which will allow more sensitive applications to use a cloud exchange.
- Cost effectiveness is another thorny issue. Public cloud storage for the enterprise isn't a race to the bottom in terms of cost; it has to be about more than just cheap capacity. Indeed, the public cloud storage market is already splitting into three categories: Cheap personal storage and backup, inexpensive storage for developers and web applications, and feature-rich enterprise-grade offerings for businesses. But cost will always be a factor, and cloud storage must prove its value. An exchange that resulted in higher prices or surprise fees to move data wouldn't be a success. This last is worth noting: Since moving storage between providers will always require time and costly bandwidth, I expect such an exchange to focus on net-new data, not migration of existing capacity to a cheaper provider.
Cloud storage is significantly different from cloud compute, and spot pricing and exchanges might make more sense for data. Indeed, many of the concerns voiced by CNET's Haff are less troubling in the storage world. But the issue of return on investment remains: Could a business cost-effectively use varying cloud storage providers? I suspect some will soon set up shop and try to find out!