News Stay informed about the latest enterprise technology news and product updates.

Tiered storage shifts focus away from disk

At Storage Decisions, users and analysts discussed creative ways to tier storage according to levels of data protection and storage services, rather than disk costs.

CHICAGO -- In response to emerging frustration with tiered storage as initially defined by the industry -- classes of disk in separate frames sorted according to cost -- users and analysts at Storage Decisions on Wednesday discussed new approaches to implementing tiers that defy the dogma.

Previous conferences and trade shows had seen users venting frustrations over tiered storage and pulling back from what was initially an enthusiastic pursuit of the process. Some initial implementations of storage tiered between Fibre Channel (FC) and SATA disks in separate arrays had actually resulted in higher storage costs and management headaches (see sidebar).

But this week found users sharing their success stories with tiering storage in novel ways, among them Harley-Davidson Motor Co.'s Andrew Madsen.

Related articles

Users get mixed results tiering storage

Users come to grips with tiered storage

Users struggle to get tiered storage right  

Five tips in five minutes: Tiered storage and ILM  

"My definition of tiered storage isn't about disk types," Madsen said. "It's about access and protection. It's saving costs, but it's not about the cost of the hardware."

Madsen said all of his storage tiers are on FC arrays from IBM. The first two tiers are both on DS8300s. The difference between them is the size and the performance of the drives -- the Tier-1 disk spindles contain 146 GB that spin at 15,000 rpm, allowing faster access to data for mission-critical databases, including Oracle. The Tier-1 segments of Harley's storage environment are also protected using RAID-10, redundant controllers and are slated to be replicated to a secondary site when it is up and running this quarter.

Meanwhile, Madsen's Tier-2 storage, located on another DS8300, contains 300 GB disks that spin at 10,000 rpm, are protected by RAID-5 and will be backed up on site. These arrays, which do not have the redundant controllers contained in his Tier-1 system, are used for development data for quality assurance and testing environments that still call for reliability but don't demand the performance of the faster 146 GB drives.

Finally, Madsen 's third tier is on a DS4800 array, which also uses 146 GB drives but with less cache than the other frames, and without dual processors.

"The difference between the 8300s and the 4800 is that if we lose processors on the 8300s, we're down ports, but the arrays won't fail," he said.

All told, Madsen estimated tiering the levels of protection and performance on his arrays had saved him approximately $500,000 this year, representing 50% of the disk budget and one fifth of his overall budget -- all without a single SATA disk implemented.

Another unorthodox approach was taken by Blackhawk Technical College, which plans to stray from the norm by bumping a FC box from Hitachi Data Systems down to Tier-2 while using an IP SAN from LeftHand Networks Inc. as primary storage.

"Because we're a college, we need to keep our costs down," said Jeff Redington, systems engineer with Blackhawk. Cost was his primary motivation for looking into iSCSI, he said. Currently, Redington said he is looking into IP SANs from LeftHand Networks, EqualLogic Inc. and Xiotech Corp., and said that though he's only gotten one price quote from LeftHand, the savings are obvious.

"HP [Hewlett-Packard Co.] gave us a quote for a midrange FC array that was $175,000 for 4 TB [terabyte)," he said. "LeftHand's quote was $100,000 for 16 TB."

Redington also said the typical concern about the performance gap between iSCSI and FC weren't as much of a concern; he reasoned that since the older Hitachi array is still running 1 Gbps FC, switching to Gigabit Ethernet wouldn't make much of a difference in his environment.

As for the other strikes against IP, Redington did say that reliability on an IP SAN is "a concern." In response, the college plans to implement the SAN on a separate IP network from its LAN; and once its final decision on a SAN vendor is made (Redington indicated he is leaning strongly toward LeftHand's system), it will run on a set of clustered disk nodes for reliability.

Meanwhile, Xiotech user Joel Grace, manager of storage systems for Houghton-Mifflin Co., subsidiary Riverside Press, said he is using FC and SATA disks in his environment for different tiers of storage, but he has bucked the conventional wisdom by sticking fast to just two tiers rather than the recommended minimum of three (in fact, in his presentations, Casey recommends users come up with between four and eight classes of data).

Grace said high-end databases reside on 15,000 rpm 146 GB FC disk on a Magnitude 3D array from Xiotech protected with RAID-10. This data is replicated to a Magnitude classic array that otherwise would have been discarded. The second tier on 7200 rpm 400 GB SATA drives, also on a Magnitude 3D, is protected with RAID-5 and is backed up to tape on site.

Grace said it had been made clear to him that at least three levels of tiered storage were the norm, but "three levels to manage would negate the cost savings of the cheaper disk," he said. By sticking to his guns, repurposing the Magnitude classic and changing data protection levels between his arrays, Grace estimated he had still saved as much as $200,000.

A disconnect between vendor and user strategies

There has been recent evidence that some vendors are beginning to catch on to the backlash against tiering classes of disk. For example, EMC Corp.'s recent Clariion refresh no longer uses SATA disk for lower tiers, instead favoring low-cost FC disk, and HP uses FATA rather than SATA in several of its midrange arrays. Another player, Compellent, takes a "none of the above view," according to spokesman Rob Davis.

"We believe data should be migrated automatically between storage tiers according to usage patterns rather than by type or application," Davis said in an email, "as is being pushed aggressively by some consultants and the industry's largest vendors who offer products and services in this area."

Meanwhile, users and analysts said on Wednesday that disconnects remain between the marketing of tiered storage products in general (or trying to circumvent them by slashing prices on high-end arrays) and how users are attempting to implement them.

Just as Grace has heard much about the benefits of a three-level system, Madsen said he has found resistance in some cases to his tiered storage plan. He said IBM has urged him to use 146 GB drives in both 8300 systems, citing their higher performance and the fact that disk is cheap.

"Cheap disk doesn't solve the problem," according to Casey. An informal poll of the audience showed 55% agreed with Casey that a significant portion (20-50%) of their data is "over-protected," regardless of the price of the hard drives used to store it.

More to the point, a slide was included in the "Tiered Storage School" presentation Wednesday morning by Mike Casey, vice president and chief technology officer of consulting firm Contoural Inc., titled "Why your incumbent vendors may not embrace true storage tiering."

"Tiered storage done right can avoid Tier-1 purchases for the next two years," according to Casey "Most vendors don't offer best-of-breed in all tiers -- this is bad news for them."

Casey also said vendors bristle at his tip that users focus further storage purchases on equipment for the low end, rather than for Tier-1.

In fact, Casey said, some of the cases in which tiered storage had been a spectacular success for his clients hadn't involved the purchase of new storage at all. One customer, whom Casey did not name, had saved some $5 million just by changing which data it used replication to protect, he said.

Casey also said much of the frustration with tiered storage so far has come as a result of the confusion over tiered storage and information lifecycle management (ILM). "ILM is data movement over time, while tiered storage is the placement of data on the most cost-effective storage for its first use," Casey explained. "ILM and archiving can take six months to implement correctly, whereas tiering storage shouldn't take more than two months."

Ultimately, he acknowledged, many users are still thrown by higher up-front costs for implementing a tiered storage system. But he urged them to take the long view.

"Tiered storage done right may actually see an incremental cost increase up front," he said. "But you can get through your next hardware refresh without purchasing any more high-end equipment -- that's where the real savings are."


Dig Deeper on Storage tiering

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.