Gordon McKemie

Mon, 14 Dec 2009 10:00:00 -0600

Two interesting articles in the press today—one discusses the impending release of EMC FAST. George Crump is speaking of “Policy-Based” storage management, here, BTW. Something that is in the Compellent labs now. My comments below….G
Why Stop At Automated Storage Tiering?

Posted by George Crump, Dec 11, 2009 11:29 AM

Automated tiering, the transparent movement of data based on activity or type, is quickly proving itself to be a hot consideration for storage managers but why stop at automated tiering? Can’t we make the entire storage ecosystem respond automatically based on environmental conditions and its available resources?

Driven in large part by storage companies and storage managers trying to decide how to best take advantage of Solid State Disk (SSD), automated tiering solutions are trying to automate the movement of hot data. EMC for example this week released 1.0 of its FAST (Fully Automated Storage Tiering). Howard Marks gives a great summary over on Network Computing. Automated tiering is not new. Compellent, 3PAR, Dataram and FalconStor have been doing something similar for a while on block storage. We have also seen companies like Storspeed and Avere offer similar solutions on NAS based systems.
Again, why stop at tiering? Data protection decisions could be automated in much the same way. Here the industry could learn from the Data Robotics Drobo which can transparently adjust protection levels based on available capacities. Enterprise storage systems in the same manor should be able to respond to the insertion of any amount of storage, classify that storage and decide how that storage can allow the current data protection method to improve. If you have enough capacity why not mirror everything initially, then downgrade to RAID 6 and then RAID 5 as capacity becomes more scarce? Of course you would want some notification or warning from the system that it is going to make these changes, but why should storage administrators have to waste time making them?
Along the same lines if you implement a second system that has spare capacity on it, why not have the primary system automatically start performing continuous data protect (CDP) of its most active volumes to the spare capacity on the secondary system? Further if they find another one of themselves on the network, maybe have them perform WAN replication. Some of today’s storage systems are essentially running on a Linux or Windows core. Why not have those systems be able to do a image dump of data to a connected tape or deduplication system?
There are downsides that need to be worked through with this level of automation, and there are going to be storage guys like me that want to have the ability to tune and tweak. For a growing number of IT professionals however, there is simply too much data to try to manage it all. The thought of a Drobo like black box for the enterprise that automatically understands the storage demands of environment and then provides the best performance and reliability based on its available resources could have strong appeal.
Track us on Twitter: http://twitter.com/storageswiss
Subscribe to our RSS feed.
George Crump is lead analyst of Storage Switzerland, an IT analyst firm focused on the storage and virtualization segments. Find Storage Switzerland's disclosure statement here.

EMC Delivers On FAST 1.0 - Call Me When v2 Is Ready
Posted by Howard Marks on December 11, 2009
This week EMC made a big splash, announcing that they're actually delivering the first version of the FAST (Fully Automated Storage Tiering). Now owners of the latest EMC kit can automatically migrate LUNs from one tier of storage to another. While that's a lot better than rocks for Christmas, it's really just a down payment on the best present ever. EMC is promising more later, and I don't even think they're keeping track of who's naughty or nice.

Of course the announcement was accompanied by the EMC bloggers all describing how wonderful the future would be when there was FAST moving data through thin volumes to deduped/compressed stores and off to federated cloud storage just like slides 11&12 on the PowerPoint deck. That was followed by press releases and blog entries from the competitors explaining how they've been doing something almost as good, or in Compellent's case, better, for years.

We all know that placing the busiest 2-5 percent of our data on SSDs would let us put most of the rest on capacity oriented SATA or SAS drives. That would save even bigger bucks than we spent on the SSDs and boost application performance. The problem is we don't know which 2 percent of our data makes up our hotspots. FAST can automatically identify the LUNs in a subsystem that are being hit the hardest and move them up to a faster, probably flash based, storage tier, and that's a good start.

Users will have to make some changes to their data management processes to get the most of FAST. First, the storage admins have to work with their DBAs and application admins to tease as much of the cold data to different volumes than the cold data. Since the first version of FAST doesn't support thinly provisioned volumes, they'll also have to stop using standard size LUNs and overprovisioning. If each of three 50GB tablespaces are allocated 250GB LUNs because 250GB is the standard LUN size for Oracle, only one will fit in 300GB of flash, but if each is allocated 75GB, they'll all fit. Of course tighter allocation means more monitoring and expanding LUNs.

On Celerra NAS systems FAST migrates individual files between tiers rather than LUNs so users like architects and other creative types that work with files will get the performance boost of having the files they're working on this week on a fast tier without the data management overhead. This could be another good reason to use NAS as to host VMware images especially if virtual server admins segregate their data onto multiple logical drives and .VMDK files.
FAST v1 puts EMC in the small pack leading the race for effective automated tiering. Compellent leads the way, since Storage Center is the only product that tracks access frequency and relocates data at the block level. EMC is now collecting the data and will do sub-LUN relocations in the next version of FAST due next year. I expect that's when we'll start seeing automatic tiering making a big impact on real users.

On the file front, Symantec's VxFS file system for Unix/Linux, part of the storage foundation bundle, can locate files based on access temperature and has just been updated to recognize flash volumes. Since VxFS is host based the high speed and low speed tiers can be on different arrays.

Comment by gmckemie402 on December 14, 2009 9:54 AM
Interesting commentary, Howard.
ILM, HSM, or storage tiering, whatever you want to call it, was crudely implemented on mainframe computing twenty five years ago when the only choices for storage media were disk and tape. Today, with more choices available (SSD, enterprise and archive disks, multiple RAID methods, virtual and automated tape) there are more environmental, performance and cost advantages for firms looking to implement this technology.
During the past five years the "theories" discussed in columns and dialogue among those willing to conjecture on this subject have given my lots of laughs. While the "experts" have developed theories for developing and planning business strategies on ways to accomplish this new bold solution to the problem of data explosion, my clients have been easily implementing an effective and powerful solution--Compellent StorageCenter.
Now that the legacy vendors, after years of criticizing virtual block data placement, have adopted the technology for their own, can someone in the press finally recognize Larry Aszmann and Phil Soran at Compellent for being visionaries? These guys just get it....
Gordon McKemie, OVSC
Note: Ohio Valley Storage Consultants is a Compellent Business Partner

Live from C-Drive: Keeping Storage Costs Low in a High-Growth Environment Comments

Gordon McKemie