Originating Author: David Cahill
The State of the State
Cloud storage gateways are physical or virtual appliances that reside on premise and provide a low friction on-ramp to the cloud without completely relinquishing control of an organization's data. Tactically, these gateways serve as an abstraction layer meant to alleviate the burden of integrating applications with cloud storage services by translating the cloud protocols (e.g. SOAP/REST) to the more familiar block- or file-based interfaces.
Strategically, these systems are disaggregating the storage controller from the disk. The brains of the system remain on premise, while disk is replaced locally with flash. The cloud serves as an integrated archive tier for non-working set data. If you fall into the camp that considers the cloud as an extension of existing IT, then the cloud storage gateway starts to look a lot like a next generation storage system.
From a feature/function standpoint the gateway vendors are building in “table stakes” enterprise services like snapshot, replication, and thin provisioning into the on-premise controller footprint. From a performance standpoint, the approaches vary based on form factor and protocol choices, but most have an SSD tier and a dynamic cache to minimize fetching data from the cloud. In instances where reads/writes require cloud access, various forms of de-duplication, compression, and WAN optimization work to minimize latency and network load (i.e. cloud provider transfer fees). From a security standpoint, these devices typically include encryption and key management handled locally. Early on, the most appealing application of gateways has been as a backup target, archive tier, or tape replacement. The more ambitious vendors are building global file system like capabilities to enable functionality like collaboration and multi-site replication.
From a business model perspective, the gateway opportunity takes data off a high ASP, 50-60% gross margin, disk system and pushes it to a smaller, lower ASP, appliance footprint which then offloads the bulk of the capacity to a commodity disk pool. Although models vary, the gateway vendor gets a capacity management license and additional revenue as its footprint grows. The cloud storage service provider gets $/gb under management and related data transfer fees. More interesting, the majority of this revenue stream falls outside the Tier 1 Storage OEMs. Moreover, early-use cases for cloud gateways are creating value by pushing out the need to purchase incremental expensive Tier 1 disk.
Backed by considerable venture funding, several emerging vendors are competing in and around this space, including CTERA, Nasuni, Nirvanix, Panzura, StorSimple and TwinStrata. Established vendors with offerings include F5 Networks and Riverbed. Cirtas was playing in this space as well before a very public unraveling back in April. Two weeks before Cirtas blew up, I listened to a very happy Cirtas customer talk about his experience at Storage Networking World in Santa Clara. And while the reasons for Cirtas’ demise are up for debate, it is my sense that its downfall was more company-specific and not necessarily an indictment on the viability of the cloud storage gateway opportunity. So rather than spend a lot of time sorting through what this means for the cloud storage gateway market, it seems more productive to put this topic aside and move on.
The Battle for the Working Set
For random I/O intensive applications (e.g. database) where cache is far less effective, gateways run into performance limitations today. Distance remains the primary inhibitor to broader tier 1 application adoption. Each of the gateway vendors has some form of intelligent tiering algorithms doing everything they can to ensure all working set data remains on premise in the local footprint. The good news here is, true transactional I/O datasets (i.e. the working set) are significantly smaller than the sequential loads. The bad news is the current storage OEMs have a massive disincentive to telling you this.
The Great Disk Subsidy
Unfortunately, large storage OEMs are still too dependent on massive disk-related revenue streams to create any sort of incentive for customers to move away from them. Even the best technology story, and related customer value proposition, will struggle to survive if it conflicts with the major OEM’s agenda. Cloud gateways, while initially positioned as an augmentation, directly conflict with this agenda.
The gateway opportunity is conducive to emerging companies because they aren’t reliant on legacy revenue streams heavily subsidized by disk sales. Over time, flash prices will continue to decline, making it more economical to have a larger and more effective local cache. As the high-performance flash footprint of these systems increases, so does the applicable workload TAM. Inevitably, these systems encroach on the legacy onsite footprint for at least tier 2/3 disk and tape replacement. In fact these systems must cannibalize that revenue stream to remain viable long-term.
No Dog In The Fight
In the past, emerging technologies in storage would rise up and disrupt a piece of the on-premise storage footprint, forcing OEMs to pay attention. The response from the OEM is usually some sort of reseller or OEM agreement to fill the void and buy time to figure out a strategy. Once they conclude that they can’t build the equivalent technology fast enough, they go out and acquire the feature company, incorporate it into their system, and continue selling more disk.
The major difference with cloud gateways compared to other features is that a large component of the future revenue opportunity here is not purely an on-premise discussion. Early vendors in the deduplication and VTL markets cannibalized tape revenue and kept it out of the hands of the large disk vendors. However that data was still maintained on-premise. Once the opportunity was big enough, OEMs could acquire this technology, build it into their systems and retain the disk revenue opportunity.
The conflict posed by cloud gateways is they are typically offloading data to a non-traditional cloud storage vendor (e.g. Amazon, Nirvanix, Rackspace, Windows Azure). Legacy storage OEMs, without a dog in the cloud fight, have a clear disincentive to promote the migration of data from on-premise to the cloud. Some will argue that these legacy vendors are playing in the cloud by powering the storage offerings of the cloud service providers. And while that may be true, this is a very different position in the supply chain than they are accustomed to. More importantly they are relinquishing customer control in the process. This shift has material implications, not only on supplier/buyer power, but also in the manner in which the equity markets value an OEM supplier versus an OEM. Short of buying a cloud storage provider and competing head with the legacy service providers (which still might be the right thing to do), the NetApp’s and EMC’s of the world are increasingly marginalized to the role of supplier in the cloud storage discussion.
We’ve seen this movie before
This whole cloud gateway idea feels an awful lot like the open systems virtual tape movement a few years back. In the case of VTL, hard disks were all of a sudden a cost-effective backup, target but backup applications didn’t know how to speak disk at the time. A bunch of companies/products were spawned to avoid, or at least delay, the burden of integrating directly to disk. The VTL appliances provided quick time-to-value, reduced backup windows and eliminated physical tape management. Deduplication was eventually added to the mix and then arguably consumed a large portion of the VTL opportunity by presenting a file system interface to applications that eventually learned to speak directly to disk. Data Domain was the big winner in this opportunity due to an elegant architecture and strong channel strategy, coupled with TTM (time-to-market) and TTV (time-to-value) leadership.
Is it different this time?
Why are cloud storage gateways any different from this VTL movement? Cloud gateways provide a familiar interface to the applications and speak cloud out the back end, with all the appropriate integrations to make it work. But like VTL and dedupe before it, doesn’t this gateway feature set get consumed into the bigger storage systems over time? History says features almost always get consumed into systems. Does the cloud in any way alter the outcome this time around? Is there any chance these appliances marginalize the legacy storage footprint by offloading all but the most random and latency-sensitive workloads? Does this cloud storage gateway eventually morph into the next generation storage system? Or do legacy storage vendors eventually build cloud APIs into their existing systems, making life much more difficult for emerging vendors.
Thin Client, Fat Cloud
Gateways are an interesting tactical opportunity, but these emerging vendors, absent owning the back end storage as well (e.g. Nirvanix), need to figure out a way to better monetize the control point they are establishing. This is where this opportunity gets interesting. While gateways are an intelligent pass-through device today, what prevents them from evolving into the next generation storage system? Over time, systems always seem to consume features in the enterprise storage space. Why couldn’t it be different this time around? Today, these gateway vendors are proposing to deliver an enterprise feature-set to a targeted subset of a customer’s data. And while use cases are initially limited, it is still early. Mobile devices, with continual advances in processing power and density, continue to evolve and marginalize PC use cases. Why can’t gateways have a similar impact on the storage market?
Amazon Web Services VP & Distinguished Engineer, James Hamilton, in a great presentation on Internet Scale Storage given at SIGMOD 2011, points to mobile phone design as the best leading indicator for future data center innovation. Mobile phones are thin, yet converged, feature-rich, devices that function as a presentation layer for a bunch of applications that are served up from some sort of cloud. These initial cloud gateways are essentially converged storage appliances serving as a presentation layer to data that resides in the cloud. Certain applications still reside locally, but as the cost of flash continues to decline, the onsite cache will become larger and cheaper on a cost/GB basis.
If this mobile phone analogy has any merit, doesn’t this also mean that the more lucrative long-term model for the cloud gateway opportunity is providing an ongoing subscription-based storage service? I have to believe that a lot of people would be interested in owning this type of annuity stream.
Enabling the Channel
Beyond a compelling technical value proposition, the other major component of the gateway story is the ability to enable the channel. VARs and resellers are struggling to figure out how to participate in the cloud discussion on an ongoing basis. Meanwhile, the Amazons of the world are not proactive enough to be relied on as a legitimate source of business.
Therefore, channel enablement here is as meaningful as providing leading technology. If the cloud storage gateway can accelerate the evolution of the VAR business model from box pusher to service provider, then this gateway concept becomes that much more powerful. Simultaneously enabling customers and arming the channel with a cloud story is a very interesting proposition. Established vendors have all sorts of business model and organizational inertia that make it difficult to tell a story like this. The unknown here is whether some of the lower ASP gateway offerings are attractive enough for the channel. To be successful it is up to the vendors to devise a model that will make the traditional channel pay attention.
Begin with the end in mind
Even if the traditional storage system eventually consumes the gateway feature-set, there is still probably room for one decent exit to whoever builds the strongest early traction with customers and the channel. This would look most similar to the deduplication market with one large exit, Data Domain, and a few smaller ones like Avamar and Diligent. The more optimistic scenario is that cloud storage gateways increasingly marginalize the on-premise storage footprint. This scenario provides room for multiple vendors to establish legitimate run-rate businesses. This would look similar to what we saw with 3Par, Compellent, Equallogic, Isilon, and Lefthand before the recent market consolidation. As a huge fan of disruptive innovation, I am cheering for the latter scenario. However, the startups in the space have their work cut out for them.
It is only a matter of time before all the major storage vendors build cloud gateway functionality into their storage systems or NAS gateways. The emerging gateway vendors need to architect for this reality today. Those that are building a primary storage system disguised as a gateway, or those that own both the gateway and the cloud, are the most interesting technologies for both customers and acquirers.
Opportunistic feature companies will capitalize on the tactical gateway opportunity. But the more sustainable business models are using cloud gateways as a Trojan horse entre into the enterprise storage discussion. A land-and-expand strategy that leverages business model agility and clean slate thinking to achieve TTM and TTV leadership is well positioned to capitalize on the general sense of urgency around participating in the cloud opportunity. As flash economics continue to improve, why can’t the future on-premise storage footprint look more like a gateway than a traditional storage system? The answer is, it probably can. The question is whether it will be the legacy vendors that evolve to this reality or whether the emerging players, building off the cloud onramp beachhead, can establish a more viable and enduring business model.
Footnotes: Dave Cahill is a Wikibon contributor and founder of Diligence Technology Advisors. Diligence is a boutique strategy consulting practice with a focus on storage, cloud computing and emerging technology opportunities. Dave can be reached via email (firstname.lastname@example.org) or twitter (@dcahill8).