With solid state drives (SSDs) moving solidly (so to speak) into the data storage mainstream, enterprise storage vendors are falling all over themselves to proclaim their superiority at managing the pricey high-performance drives.
EMC’s (NASDAQ: EMC) announcement last month of its fully automated storage tiering (FAST) capability sparked a flurry of responses from the likes of Compellent (NYSE: CML), 3Par (NYSE: PAR) and BlueARC about the tiering and SSD-friendly features of their storage systems.
So let’s take a look at the market for flash solid state drive tiering, who is offering what, and where the technology is heading.
Storage tiering is far from new. Hierarchical storage management (HSM) has been around for decades, and information lifecycle management (ILM) gave the concept a new lease on life in recent years. The basic idea is to control storage capacity costs while maintaining appropriate levels of performance.
Many of the older systems, however, required manual data movement. The latest offerings eliminate the time required by storage managers to move data around. They include features such as set-and-forget policy management and tracking of file-access frequency, which determines the appropriate tier. For example, dormant or infrequently accessed files can be migrated to SATAdrives, and frequently accessed files can be stored on or moved to the fastest available medium.
SSD has given further impetus to this market and accelerated adoption. Rather than using Fibre Channel (FC) disk for top priority files in Tier 1, SSD is being harnessed as a much smaller Tier 0, augmented by a large SATA repository for the bulk of data. FC might sit in between as an intermediate tier. This arrangement provides many benefits, including a cut in power consumption.
“SSD as a technology does not require as much power as spinning disks,” said Gerard Sample, senior manager of product marketing at BlueArc. “And a higher percentage of energy-efficient drives, such as SATA, will improve overall system power consumption.”
Some vendors simply throw a few SSDs into the mix as a performance accelerator overlay for an entire system. This method reduces the number of FC or SAS drives needed to deliver acceptable system performance. But as time progresses, management of SSDs in a tiered architecture is becoming far more sophisticated.
“Consider the concept of a multi-tiered file system where metadata tables are transparently placed on SSD and file data on a SATA tier,” said Sample. “Given that response times are often predicated by metadata lookup processes, this method results in significant performance improvements in addition to cost benefits.”
File Access Patterns, Applications Determine Storage Type
BlueArc is one of the storage vendors that started down this road a few years ago, as did Compellent and 3PAR. BlueArc has formed an SSD partnership with Texas Memory Systems. The company’s Intelligent Tiered Storage architecture comprises Data Migration and Dynamic Read Caching capabilities. These features allow administrators to set policies based on common file attributes or access patterns to ensure that data is automatically migrated to the appropriate tier of storage, including SSD or any type of disk drive.
Dynamic Read Caching instantly copies files to an SSD tier for use across physical or virtual storage servers. When a file residing on a slower tier is accessed, a background copy is made to a high-performance cache tier, ensuring that subsequent reads will be delivered at maximum speed. As the SSD tier fills up, the least frequently read file is dropped. Writes to the files are passed through to the back-end disk to ensure that replication and other file services are not interrupted. In addition to integrated storage tiers, files can be cached from external devices within the namespace, including deduplication devices or stranded filers accessible via NFS. The company’s tiering tools operate with its own hardware products, as well as arrays from Hitachi Data Systems (HDS), Texas Memory Systems and Data Direct Networks.
BlueArc takes a granular approach whereby a tier could be based on a drive type, group ownership, the day of the week, business function or the application feeding that storage tier. It can also extend beyond a single system to the arrays of other network-attached storage (NAS) vendors.
RAID, Block and Platter Data Placement
Compellent is another vendor vying for automated tiering attention.
“SSDs are a natural fit for tiered storage environments, as their performance benefits are best leveraged for enterprise applications that require high IOPS, such as transactional databases, rather than stale or less frequently accessed data,” said Bob Fine, director of product marketing at Compellent. “The key to making tiered storage easy to manage, low-cost and efficient is the level of automation and system intelligence.”
Compellent also advocates automating tiered storage at a granular level, as it enables companies to avoid having to manually move volumes between tiers or engage expensive professional services. This is done via its Data Progression software, which can move individual blocks within the volume between tiers. It recognizes different drives as they’re installed and assigns them to the appropriate tier based on pre-defined profiles.
“Compellent was the first to offer an automated tiered storage solution, Data Progression, in its SAN, back in 2005,” said Fine. “Since only active data is stored on the drives, customers only need to invest in as few as three SSDs, instead of the bricks of eight or 10 SSDs required by some competitive systems.”
Blocks are automatically migrated based on frequency of access from the inner to the outer tracks of every disk drive — the fastest part of a drive is typically the outermost edge. Some vendors put entire volumes on the outer tracks, which can negate the efficiency advantages of automated tiering, Compellent says. Data Progression also tiers data between RAID volumes. In a single tier of FC storage, for instance, you can migrate the inactive blocks from RAID 10 to RAID 5 to save on disk costs.
3PAR, meanwhile, offers its Policy Advisor software that adds further policy management and automation capabilities to its Dynamic Optimization for 3PAR InServ utility storage servers. It analyzes how virtual volumes use physical disk space and makes automatic adjustments to ensure optimal volume distribution and storage tiering.
FAST Times at Hopkinton High
Others in the game include Sun’s (NASDAQ: JAVA) ZFS file system, one of the earliest SSD tiering entries, HDS, HP (NYSE: HPQ) and, of course, EMC.
The first version of EMC FAST operates with Clariion, Symmetrix and Celerra storage arrays. What most of the competitors are pointing out is that the first version only functions at the LUNand not the block level. Further, rivals point out that FAST locks devices when data is being swapped around. EMC plans to fix much of this by version two, which is due later this year. So why not wait till then for the initial FAST release?
Competitors certainly have their opinions.
“The primary purpose for the FAST initiative is to sell more flash SSD drives,” said Sample.
“FAST and other solutions only have the capability to move entire volumes of data between tiers, while Compellent’s Data Progression moves data at 512KB blocks, regardless of the storage volume, spindle speed or disk type,” said Fine.
Whatever the others say about EMC, however, it appears that FAST has gotten their attention. While the others have achieved modest adoption rates, EMC’s clout and market domination could well transform automated tiering into a mainstream technology.
Gordon Haff, an analyst at Illuminata, believes that FC drives will begin to be replaced over time by a combination of solid state and SATA and SASdisk drives. And EMC is likely to be heavily involved in that transition, especially once FAST v2 comes out.
“EMC told me in a recent briefing that their goal in FAST version 2, slated for mid-2010, is to largely mask platform differences from users using management and other administrative interfaces,” said Haff. “FAST v2 will enable the relocation of blocks of under 1 megabyte.”
Follow Enterprise Storage Forum on Twitter