Unlike many of my fellow bloggers, it seems, I have been engaging with consumers — nearly a hundred in the last two weeks — and gathering their insights about this whole tempest in a teapot called on-array tiering. That’s code for 1) establishing a tier of Flash SSD in a conventional disk array, 2) adding low cost capacity drives to the same rig as higher cost performance disk, then 3) adding more code to an already bloated controller to establish a rudimentary hierarchical storage management (HSM) process between Flash, performance disk and capacity disk. The folks I am talking to have no use for it. They see it, as do I, as more vendor foo.
I was included in a blanket post over at StorageRap covering this subject where I was dismissed as a “contrarian” who “supported” NetApp CEO Georgens’ position on tiering. The discussion there in the comments devolved into a geekfest over the merits of NetApp WAFL versus other vendor tiering approaches that, I guess, someone would want to read if they are in need of an espresso shot of storage technology dillettantery. I prefer the real world, personally.
I know about NetApp architecture — write spoofing, NVRAM caching, WAFL and RAID N — and how the scheme surmounts write penalties. I also know the arguments of the on-array tierers and how they think that a collection of overpriced hardware and software will someday make up for its obscene expense with greater administrative labor cost savings.
Frankly, and this IS the contrarian in me, neither I nor the hundreds of IT folks I have spoken to in the past eight days, give a rat’s butt about either one. The issue here goes deeper than hardware innovations. It goes to simple costs in the real world.
NetApp stuff is very expensive. Made moreso by aftersale warranty and maintenance costs — and the non-transferability of licenses that give their gear exactly zero resale value. While folks I talk to tend to like ONTAP, they don’t like the sticker, which is why little plays that aren’t getting much attention right now, like SimplStor, are beginning to resonate with even big name NAS consumers.
Before the anti-NetApp pro-tiering folks get all uppity, you should know that very few of these IT mavens believe anything the on-array tiering folks are saying. They doubt that Flash SSD has a sustainable value proposition under their workload. To a one, they have had really bad experiences with automated tiering via HSM software because, frankly, automated HSM storage movements are viewed as inherently dangerous.
A key issue that none of the vendor advocates-qua-expert bloggers seems to be discussing in their rants around the wonderful world of HSM is what happens when you move some data that actually needs to be restored to its previous space to get work done — only the space is now used by something else. This happens all the time, especially when date last modified or simple watermarking is used as the trigger for movement in the first place. One guy said that he had problems when data was moved that hadn’t been touched in three months, but was nonetheless required to do end of quarter or end of year reporting. You get the idea.
Truth be told, you need to have a clue about your data and its BUSINESS CONTEXT (aka class) in order to make tiering work at all. That’s an old beef of mainframers with respect to DFHSM, by the way: near perfect understanding of data class is required for DFHSM to deliver its promised value and the lack of flexibility in DFHSM class definition is a bane of existence for nearly everyone who has used it.
So, you need to know your data. That’s before you start tinkering with HSM policy definition. Yet, the on-array tiering vendors are claiming that their automation eliminates the need to do the heavy lifting of data classification. Without it, I expect to be writing case studies of EPIC FAILURES of on-array tiering at some point in the future because I know that EMC or someone will sell a few of these rigs to those who simply outsource their thinking process to their vendor.
A key thing I heard at FujiFilm’s Global IT Executive Summit in Austin last week was that even technologies like de-duplication were requiring their users to go back to the data and to understand it better, lest the Data Domain rig be used inefficiently.
They don’t really give a tinker’s damn about de-dupe or on-array tiering or thin provisioning on the controller. They see these features as toys offered by vendors to make their boxes of spinning rust more interesting and as cost accelerators of the overall solution.
I can’t tell you how many IT executives offered comments that distilled into a simple wish: “give me a rig that is basic, affordable, and manageable with all my other stuff through a single pane of glass.”
They like the line of Hu Yoshida at HDS, Ziya Aral at DataCore, ReiJane Huai at FalconStor and others who argue that all of the value add functionality needs to be abstracted away from the commodity disk and offered as a set of optional services to be turned on or off as needed by different data workload.
There was general agreement that the software on bloated controllers is what fails — far more often than the hardware it controls.
Plus, they don’t like that vendors are force marching them in the direction of “smart array controllers” that are being used to jack up the cost of the commodity disk in the array in order to line the pockets of the vendor.
They perceive, correctly I think, that EMC, NetApp, et al., regard their own top line growth as inherently more important to them than the customer’s top line growth and design equipment to lock customers in while locking competitors out — thus compromising any hope of cross-platform management. Lack of common coherent management of infrastructure is clearly a cost of ownership driver that none of them — irregardless of size — can afford.
Look, if we want to do HSM, there are marvelous software solutions that we can use that enable us to buy just the bare storage at rock bottom prices. For more resiliency, we can look at some of the brand name and off-brand storage rigs whose engineers have done great things with RAID, RAGS, on-array drive remediation, etc., to ensure hardware availability while supporting common cross-box management. Doing tiering on the rig creates a stovepipe. Good for the vendor, bad for the consumer.
I usually don’t respond to too many fellow bloggers because they form a little club of insiders who like to debate angels on heads of pins — apparently to the benefit of their undisclosed vendor sponsors, if recent bruhaha’s on this issue in the trade press are factual. I really don’t like it when one blogger disparages another as a “storage blogging wannabe” — a reference made by StorageRap to describe Bruce Kornfeld over at Compellent. Bruce is a VP of Marketing for the company and a likeable guy with his own insights born of his own interactions with his engineers and customers. Why is he a wannabe? Because Farley said so?
My purpose in blogging originally was not to be a contrarian. Back in May 2005, I started this blog in large part to store my notes for articles and books in a manner that made them accessible to me while traveling to my client sites around the world. It morphed over time to provide an outlet for venting my bile or showing off my questionable artistic skills with PhotoShop or promoting some ideas I had been noodling over. I have found it enormously valuable to have feedback from readers, both engineers who could fact check me, and users who could tell me their experiences. The experience has been very rewarding and is, in my humble opinion, still an adventure. I am a proud storage blogging wannabe — not an expert, I guess, like StorageRap.
You must log in to post a comment.