Storage 2008: Promises Delivered and Postponed, Part 2
Even before green got hot, it was apparent to many IT mavens that storage was actually consuming the most power and generating the most heat in their shops.
Green IT became a buzz phrase in late 2007, with vendors practically tripping over themselves to re-cast their wares in the trendy new color. With so many industry cliques forming to develop metrics to analyze and, ultimately, to drive down the power consumption of their gear, one some major breakthroughs in 2008 would have been expected to attack the kilowatts consumed and BTUs generated by IT operations.
Server consolidation via blade technology and virtualization came to the fore almost immediately, attacking server hardware -- that a study by Lawrence Livermore scientist Jonathan Koomey identified as the key power pig in the data center. This study was the darling of both industry and government in 2008, laying the foundation in the DOE’s 2008 Energy Star program, despite its obvious flaws.
Koomey focused exclusively on server power consumption while ignoring power demands of network and storage gear, used mistaken IDC projections of server deployment growth rates (dating back to 2005), and based the study's calculations on the rarely accurate power-consumption estimates of server manufacturers. Discussions of his alarming findings, however, succeeded in uniting the analysis of facilities and IT budgets in many companies, heretofore often treated in isolation of each other.
Even before green got hot, it was apparent to many IT mavens that storage -- and not servers -- was actually consuming the most power and generating the most heat in their shops as a function of poor capacity allocation management and even worse utilization efficiency. The former reflected a deficit in capacity management tools and failure to deploy effective storage resource management (SRM) technology. The latter reflected a near-total lack of effective data management, including data classification, hygiene, and intelligent archive.
Storage hardware vendors did what storage hardware vendors always do: they promoted solutions for operational greening requirements that involved plugging more gear into the wall. NetApp led the charge initially with a recommendation that consumers consider re-driving their NAS appliances with bigger hard disks. That way, for the same carbon footprint, companies would grow their capacity to store more data.
The catch was that larger capacity drives required an upgrade of the RAIDing scheme used effectively to protect smaller drives. RAID 4 (in NetApp’s case) and RAID 5 (in the case of many other arrays) were considered too slow to rebuild a failed high-capacity disk. RAIDn (in NetApp’s case) and RAID 6 and other non-standard RAID schemes needed to be substituted, required replacing backplanes and other key components of the NAS head. In effect, array re-driving translated to a rip-and-replace hardware strategy that fewer companies could afford in the current economy.
Other hardware vendors seized upon “value-add” technologies such as block de-duplication, thin provisioning, and MAID to enhance capacity allocation efficiency or to spin down disk containing rarely accessed data. Not surprisingly, substituting green media such as tape or optical for spinning rust was rarely discussed at storage vendor-driven green IT conferences. Instead, vendors offered that filling racks completely with trays of drives would reduce hot spots that drive data center power and cooling requirements.
At the very time that EMC was proposing more disk to solve power issues (and working overtime to add the “value-add” features enumerated above to their gear, which was being replaced in some shops by new players in storage such as Compellent, 3PAR, and COPAN Systems), they were also running a well-funded campaign to destroy the perceived efficacy of optical storage. Arguably, the “shatter the platter” marketing push hit its target when the leader in the optical space, Plasmon, went under in 2008.
Even the Storage Networking Industry Association (SNIA), while endeavoring to wrap itself in the green flag, carried the party line of the leading disk array manufacturers. The organization published an unscientific survey research paper early in the year suggesting that archive was such a difficult undertaking, the only way to cope with the data deluge (the root cause of storage capacity growth) was to eschew archive altogether and throw more disk at the problem. Paralleling the talking doll of a few years ago, the message from SNIA was simply, “Archive is hard. Let’s go to the mall.”
With the exception of Spectra Logic, the lack of noise from tape-library product vendors about Green IT was enough to make you check your hearing-aid battery. Sun Microsystems, having acquired the considerable tape (and disk) technology assets of STK, seemed unable to figure out how to market tape at all. In distributed systems tape, a not-impartial analyst’s endorsement of tape commissioned by the Linear Tape Open (LTO) consortium mid-year garnered more criticism -- even from tape enthusiasts -- than it did cogent assessments of the continued relevance of the technology, especially in terms of its energy conservation value.
Just when we had all but given up on green IT, two things happened. First, at CA World in Las Vegas several weeks ago, I had a conversation with an old friend who has recently become responsible for storage strategy for one of the largest grocery store chains in the United States. He said that he was working in earnest to deploy Blu-Ray optical disk as a target for archival and infrequently accessed data.
This person said that the idea had drawn a lot of fire from the other members of his strategy group, many of whom were intent to follow the lead of their preferred vendors to add more high capacity SATA arrays to store the bits. A few were promoting tape as the alternative. He challenged them to show him any sort of economic model that would provide the capacity and media durability of optical, properly deployed in libraries and not exposed to environmental media degradation factors such as ultraviolet light. His emphasis was on metrics: cost per GB from a systemic perspective including energy costs. No one could provide a plausible alternative. Needless to say, a representative from SONY at the show was delighted to make his acquaintance.
The other notable event was a conversation I had with Hubbert Smith, formerly a marketing director with Western Digital and now with disk maker Samsung. Smith is very much engaged on the green issue, but does not ignore storage performance requirements for day-to-day IT operations. He has come up with some remarkable metrics of his own to demonstrate the power, performance, and reliability benefits accrued to building tiered architecture with non-volatile, multi-layer, flash memory as tier one and SATA disk as tier two.
Placing about one-tenth of active data on flash and the other nine-tenths on 1TB SATA, his calculations show significant reductions in the costs of storage -- both in terms of fixed assets and energy consumption. The case study he provided showed a typical 60-drive Fibre Channel or SAS array versus a hybrid Flash/SATA array of comparable capacity. The former was outperformed by the latter on an IOPS-per-dollar and GB-per-dollar basis, which was expected. Smith went further to consider power consumption and demonstrated that on an IOPS per watt and GB per watt, the alternative architecture blew the high-end spinning rust out of the water.
I will discuss Smith’s views in detail in a future column. For now, your comments are welcome: firstname.lastname@example.org.