Automated Storage Tiering: EMC Is Moving FAST!

EMC has just launched the maiden version of its FAST (Fully Automated Storage Tiering) software that will run on all of its main storage platforms, Symmetrix, CLARiiON and Celerra. FAST addresses a key question for modern storage systems: in what tier should the data be placed within an array? That issue has been complicated with the recent addition of solid state disks (SSD) in the form of flash memory drives (say Tier 0) in addition to Fibre Channel (FC) drives (Tier 1) and high-capacity SATA

David Hill

December 15, 2009

7 Min Read
Network Computing logo

EMC has just launched the maiden version of its FAST (Fully Automated Storage Tiering) software that will run on all of its main storage platforms, Symmetrix, CLARiiON and Celerra. FAST addresses a key question for modern storage systems: in what tier should the data be placed within an array? That issue has been complicated with the recent addition of solid state disks (SSD) in the form of flash memory drives (say Tier 0) in addition to Fibre Channel (FC) drives (Tier 1) and high-capacity SATA drives (Tier 2). Note that practically speaking, the SATA drives are not an active archive as the placement of the data is at the block level and not at the file level, such as with Centera CAS (content-addressable storage). However, the Celerra FAST can move to locations inside or outside of an array—such as another Celerra, Centera or ATMOS—for archive purposes.

Placement of data is not obvious. Tier 0 offers the highest performance, but at the highest per byte cost although it has a cost advantage over FC in terms of $ per I/O. Tier 2 costs the least but delivers the slowest performance. Tier 1 is in the middle—lower performance and cost than Tier 0, but higher cost and better performance than Tier 2.

Getting the right answer to the question of data placement on storage tiers is very important.  Applications which need the best performance, say, for decision-making, revenue production, or customer satisfaction reasons, can benefit from the optimum placement of selected data on higher performance drives. Less business critical application data can still meet quality of service (QoS) requirements by residing on more cost effective drives. Getting the proper balance saves storage expenses and reduces energy requirements. Thus the net goal is to simultaneously raise performance service levels while at the same time lowering costs. These are the kinds of tradeoffs data center managers would like to have every day.

In and of themselves, applications know nothing about tiering. Instead, they create, read, update and delete data, but knowing where they write to and read from is beyond their purview. Manual placement of data is possible, such as for a database, but that process tends to be static or even reactive. What's preferable was a dynamic approach to placing data, wherein different pieces of data in an application could move to different tiers when and as appropriate. That is what EMC's FAST accomplishes. FAST automates the movement of data to the appropriate tier and the application is none the wiser. In a FAST environment, the data is transparent to the application at any particular time, and also to the user of the application.

Note that the static placement of data is difficult enough in a non-virtualized environment, but is totally untenable in a virtualized environment. Not only do the applications on virtual machines move hither and yon as deemed necessary, but the physical array on which the data resides may shift, for instance, in a private cloud arrangement from an internal array to a remote third party external array without advanced warning. A dynamic process requires automation in order to keep things consistent, accurate and complete as well as timely.From EMC's perspective, FAST is a fundamental building block in this whole process and an essential component in moving to IT as a service, a part of EMC's broader strategy of moving customers from a physical data center to a private cloud. IT as a service requires a flexible, self-managed infrastructure. To succeed, that effort requires numerous, seamless automated processes. From a storage perspective, that requires the automation of the movement of data among tiers of storage—hence the need for FAST.

EMC's goal for FAST is to get the right data to the right place at the right time with a process that is both policy-driven and data driven. Over time, the activities of scores of storage devices are collected and analyzed, resulting in policies being put in place. These FAST policies link storage "types" and "groups." A storage type is a shared, provisioned storage resource that contains a combination of storage media, say flash, FC and SATA devices. A storage group is typically an application, such as Oracle or Exchange, that interacts with storage types. FAST, then, is automation that dynamically monitors and relocates logical unit numbers (LUNs) based upon the particular needs of an application-dedicated storage group.
 
FAST does this first by monitoring I/O activity, such as I/O size, read density, and IOPS, at the LUN level for a period of time, the length of which is highly variable, but could be days. After enough information has been collected, FAST analyzes the data to identify candidate LUNs for relocation. Then, under complete control of the IT administrator, FAST automates the migration of the data to another tier. If the original LUN was designated as "hot," meaning that it would benefit from greater performance, the migration would be to a storage tier that can support that requirement. If the original LUN was designated as "cold," the data would be migrated to a more cost-effective storage tier whose performance should still prove adequate for the data from the original LUN.

By the way, this is not a one-time process. FAST can continue its analysis at time intervals selected by the user, such as once a day, so FAST can continue to tune and tweak the placement of data based on current user needs. Obviously, this could not reasonably be done manually, as coming back in and constantly resetting the system as conditions change would seem to be an onerous—if not nearly impossible—task.
 
EMC described use cases that were demonstrated at this year's VM World that demonstrated both performance and cost benefits. In the first, replacing 4 percent of the system's capacity with flash drives, an 80 percent busy system with only FC drives changed to only 25 percent busy. In addition, there was a 2.5X decrease in response time performance from 15 milliseconds to 6 milliseconds.  In the second, a balanced mix of flash, FC and SATA drives delivered a 27 percent lower total cost of ownership over an all FC disk environment.
 
Just a couple of housekeeping notes. EMC announced that FAST capabilities will become generally available in December 2009 for the company's Symmetrix V-Max, CLARiiON CX4, and Celerra NS product lines. Also, FAST is integrated into the Symmetrix Management Console and uses a wizard for ease of use that is especially useful in virtualized storage environments. On the CLARiiON side, FAST is a host-based CLI utility that works with Navisphere Analyzer reports. On the Celerra side, FAST is an API-based solution.

EMC, along with partner Cisco, are among the leading vendors that have a strong vision of the future. Making any and all of those visions a reality will require automation, and in the IT world at least, that means policy-driven software. The company's new FAST solution is a big step forward on the software management side, as optimal (or at least close to it) allocation of data to different tiers of storage within a single array is important from both a performance and cost perspective.

While FAST should provide useful functionality for many customers, it is still a version one product. That's not bad, but by definition no version one product is completely "finished."  Some unmet needs can be anticipated at this time, such as the need to identify data that should be moved at a lower level of granularity than the LUN level and EMC has pre-announced their intention to deliver sub-LUN level granularity in the second half of 2010. In any event, FAST is a very positive step in automating the migration of data to different tiers of storage for which many customers will benefit today as they continue on their path towards EMC's vision of the private cloud.
 

About the Author(s)

SUBSCRIBE TO OUR NEWSLETTER
Stay informed! Sign up to get expert advice and insight delivered direct to your inbox

You May Also Like


More Insights