Companies leave data on primary silos well beyond the time when they should migrate it to a less expensive tier. This is one of the few truisms on which all storage vendors seem to agree.
Of course, that agreement fades away when it comes time to define how to address the problem. Every vendor offers its own proprietary approach to facilitating allocation between storage tiers, and this variety poses one of the most formidable challenges storage administrators face as data volumes continue to increase drastically year after year.
One approach that a number of vendors are taking is to inject their products with the capability to assign aging criteria to business data and to automatically move "old" data to a different tier according to easily defined policies.
Compellent was one of the early vendors to include automatic data-migration capabilities in its products. I currently have one of Compellent's SANs in my lab for evaluation. In addition to management software and enclosures for supporting various storage tiers, it offers FC (Fibre Channel) and iSCSI connectivity; thin provisioning; choice of multiple disk drives architectures and RAID levels.
Compellent recently began shipping Version 4.0 of its Storage Center management suite. The new version includes, among others, Fast Track , a new feature that take advantage of the different access speed between the tracks of a physical disk drive.
Although not completely new, Fast Track is interesting because it creates tiers inside the disk drive in order to assign the most responsive areas of each device to the most demanding applications. Pillar Data offers a similar feature in its products. But what I want to focus on is the most traditional move of data across tiered storage -- what Compellent calls Data Progression.
For my Compellent test, I have three 3U enclosures, including one controller and two storage arrays -- one with FC and the other with SATA drives. The controller manages the connections between servers and storage arrays and hosts the management software. Compellent SANs support FC on both the front end and the back end, but because performance-testing was not an objective at this time, I decided to implement iSCSI links for my servers, leaving FC as a path between the controller and each storage enclosure.
This is what my Compellent storage environment looked like from the Storage Center GUI after a quick installation. As shown in that image, the available storage is already split in two segments, with the 15K RPM drives of the FC enclosure assigned to tier one and the comparatively slower SATA drives making up tier three. Tier two is empty in my basic setup but could be filled with an intermediate set of drives -- 10K RPM enclosures, for example.
How do you take advantage of those tiers? It starts by defining a volume: In addition to parameters such as the space to use and the number and frequency of snapshots, you can tell Storage Center's volume creation wizard how to move the data stored in that volume across tiers. Remember the Fast Track feature? For data that needs the shortest possible response time, just assign "fast" rather than "standard" track when creating the volume.
The nice touch that Compellent's Data Progression brings to tiered storage is that you don't need to analyze data or move the volume somewhere else: Behind the scenes, the system will keep track of the passing of time and will automatically migrate aging blocks to a lower tier. Should an access pattern change (say, blocks that were not accessed in months are now touched daily), the system will automatically revert the flow, moving those blocks back to a higher tier.
Controlling those flows is, you guess it, another management option. Using these settings, data that has not been touched for 12 days will be moved to a lower tier, but as I left a 4:1 ratio between down flow and upward flow, that data will be moved back if it is accessed three days in a row. Obviously, you can change those numbers to what works better for each business application.
Speaking of which, it's worth noting that from an application perspective those blocks may have never moved. The only difference could be, if ever noticeable, a slightly longer access time, milliseconds perhaps, when those data blocks are on a slower tier.
I have to let time pass to see the effect those parameters have on my volume. But I can already say that, although it may not respond to every requirement, Data Progression is one of the most effective tiered storage management solutions I have seen. It may not give you a way to classify data, say, for compliance, but it's an easy, reliable way to reduce the clutter on your primary storage devices while making sure business access is not slowed down in the process.
No comments:
Post a Comment