Almost a decade ago I ran into the 3PAR team at a trade show. I was there with my previous company, Archivas. We were showcasing the capabilities of our brand new object-store: the Archivas Storage Cluster. Based on the same distributed computing principles that had allowed companies like Amazon and Akamai to scale to petabytes, our system ran on inexpensive commodity hardware — Intel-compatible servers interconnected with plain Ethernet. The object-store was a pure software concept that still managed to scale gracefully from a terabyte to petabytes by self-organizing into larger and larger clusters. The 3PAR team was demonstrating their T Series InServ Storage Server, an impressive rack of controllers that looked like it could crush a software guy by just blinking. It was hardware. It was the old world of hard drives, blocks and LUN reservations. It seemed to me ludicrous to compete with EMC in mid-range storage arrays where CLARiiON ruled supreme. Boy, was I wrong. I was approximately $2 billion wrong.
By the time HP and Dell are done ratcheting up their bids for 3PAR the purchase price may be close $30/share or about 10 times 3PAR’s annual revenue. The rational way to justify such a large revenue multiple is the belief that 3PAR will be an asset of strategic importance. Hence, this is not a bet about the present. It’s a bet on the future. To understand the future, it helps to look back at the market forces that gave rise to the storage industry and are still fueling its development today.
Industry standards, like Fiber Channel and later iSCSI, broke the vendor lock-in that had allowed companies like IBM to charge astronomical prices for proprietary computer systems. It gradually became evident that storage was harder to switch than servers and so it became the foundation for IT. Servers came and went but storage was forever. The shift in importance from servers to storage gave rise in the 90s to pure storage powerhouses like EMC, NetApp and HDS. While giants like IBM, HP and Sun struggled to find their footing in a world where IOPS mattered more than MIPS, the pure storage OEMs raced ahead by combining relatively cheap, commodity hard drives into the fastest, most reliable storage systems in the world. Customers gladly paid a premium to build their IT towers on high quality storage.
By the turn of the millennium, anyone who was involved in storage could feel that data growth, more than any other single factor, would fuel the next wave in storage. Data was accumulating at an impressive 60% compound annual rate as the digital revolution absorbed every business process in sight. Exponential growth generated a bonanza for the OEMs that had established themselves as leaders a decade earlier. Customers, however, were less and less willing to pay exorbitant amounts of money for storage that seemed to be getting out of control. Customers also felt that these high quality storage systems had not been designed for explosive growth. The user interfaces were hardly usable and required constant visits from the vendors. Information lifecycle and similarly vacuous marketing bumper sticker technologies were hurried into the market with little success. To be fair, very few things are designed to withstand a decade’s worth of exponential growth.
Here is what 3PAR did right. They took the old technology of virtualizing blocks to its limit by producing a highly symmetrical, modular design that could scale in a mesh. It was CLARiiON but redesigned for growth. If you want something to scale, the first step is to figure out how to build it into a mesh because, otherwise, any central point of control will eventually become the bottleneck. Then they improved caching to make sure that any one component in the array could fail without compromising the performance of the entire system. They were in essence introducing some of the high-end features of the EMC Symmetrix DMX but at a lower price point and, most importantly, designed to grow incrementally with a customer’s storage needs: utility storage.
Data growth is not showing any signs of a slow down and that is driving the latest string of acquisitions as the OEMs strive to upgrade their technology arsenal. Data Domain had EMC slugging it out with NetApp for their data deduplication technology to the tune of $2.4 billion. EqualLogic and LeftHand were acquired in quick succession, by Dell and HP respectively, for their highly modular, low-cost, mid-range iSCSI arrays.
Data growth is here to stay. A decade ago, I underestimated how much pain customers would be in and just how quickly data growth would render the best technologies of the day inadequate. I also underestimated the importance of using a legacy interface. Fiber channel can cause much pain when it comes to managing growth, but it is one of a handful of standards used to provision storage. The object-stores relied on a REST APl, widely used in the Internet, but almost never used between servers and storage. It has been the job of the next generation of technology companies, like Nasuni, to develop specialized gateway products to put a native storage API on the object-stores, now better known as cloud storage.
At the current rate, data will grow more than tenfold in the coming decade. It is hard to design for this kind of growth. Vendors like 3PAR have won by pushing traditional array technology to its limit. While advanced array technology has undoubtedly helped, it fails to address the root cause of the problem. Files constitute the lion’s share of business data. Regardless of how sophisticated the underlying array technology may be, those files still need to be stored in a file system that can handle the growth, backed up and moved off-site for data protection. This is why the large Internet storage systems run by Google, Amazon and others, do not use storage arrays but instead developed their own object-stores.
I cannot help thinking that these big buys by the OEMs are akin to Polaroid doubling up on better chemistry just as customers were asking for simple digital cameras. The infrastructure of the Internet has shown that it can absorb exponential growth with ease while remaining economically viable. Cloud storage providers inherit their scalable and low cost architectures directly from the Internet. Customers still needing relief from the pains of data growth would be wise to look up, beyond the blinking lights, and into the vast potential of the cloud.