Taking charge of VM allocation, troubleshooting methods
A comprehensive collection of articles, videos and more, hand-picked by our editors
The major airlines are masters of thin provisioning; they just call it overbooking. Using the extensive data they have about every flight and its occupancy, they can pretty accurately predict how many passengers with reservations will be no-shows. So, they overbook. That we don't actually encounter overbooked flights all that often is a testament to how good their data and predictive algorithms are.
Thin provisioning is a form of overbooking. With thin provisioning, you provision a certain amount of storage capacity while actually allocating less disk capacity than you provisioned for. The application thinks it has a certain amount of capacity available to it, when it actually has less.
"Many companies are concerned about thin provisioning because they think it could get them into a difficult predicament," said Mark Teter, chief technology officer at the Denver-based Advanced Systems Group. However, if you know your applications, users, data and usage history, it's unlikely you'll encounter problems from this type of storage overbooking.
Still, if you miscalculate or something unexpected happens, you may find yourself hitting capacity utilization thresholds. If this happens and you're forced to quickly acquire and deploy more disk capacity, it still isn't a fatal problem. The capacity has already been provisioned; usually, you just need to plug in extra disk.
Let's take a look at a few things you can do to ensure thin provisioning goes smoothly:
1. Know your data
"Airlines get away with [overbooking] because they have such a detailed history of passenger behavior. They know what to expect for each flight based on different circumstances. You can do the same with storage," said Greg Schulz, founder and senior analyst at Stillwater, Minn.-based StorageIO Group.
To take advantage of thin provisioning, you have to know your storage usage and trends. "That means knowing if your data is static or changing. You also need to understand the level of I/O activity and I/O patterns," Schulz said.
In effect, "thin provisioning presents an abstracted or virtualized view to servers and applications of how much storage has been allocated," Schulz explained. In this way, thin provisioning allows storage space that has been allocated to various applications or servers but not actually consumed to be shared. Given that so much storage capacity is vastly underutilized anyway, storage administrators who understand the data and usage patterns can feel confident practicing thin provisioning.
2. Focus on file systems first
"You can be selective in where you use thin provisioning. Use it first for file services like CIFS and NFS. File storage usually is over-allocated anyway, making thin provisioning easier and less exacting," Advanced Systems Group's Teter said. With thin provisioning, a company should be able to reduce the actual allocation of disk for file storage by 20%- 30%.
Thin provisioning is less necessary and more complicated when applied to databases and structured data. Here, the database applications like to tightly specify storage allocation, and managers are rightfully cautious about disturbing carefully tuned database applications by applying thin provisioning. It's no different from IT managers who are reluctant to put mission-critical transaction databases on virtualized servers until the technology matures.
3. Leverage all your spindles for better performance
Thin provisioning can actually improve storage performance if you create thinly provisioned virtual volumes that cross as many disk spindles as possible, Teter suggested. As a general rule, the more disk spindles handling the storage request, the higher the performance. With thin provisioning, you consume less physical capacity while getting better performance.
4. Use in conjunction with storage resource management (SRM) and data deduplication
"Thin provisioning is not a product on its own, but a feature," StorageIO Group's Schulz said. Like storage resource management and data deduplication, thin provisioning is being built into storage products in various ways. It should be combined with data deduplication and SRM as part of an overall storage management strategy. Storage resource management will provide usage data that enables the organization to thinly provision the storage better. When combined with data deduplication, the organization stands to increase utilization of existing storage while reducing the need for additional storage capacity.
3PAR Inc. and DataCore Software Corp. staked out positions in the thin provisioning market early on by selling data storage arrays and software that made extensive use of storage virtualization when allocating capacity.
However, storage vendors like Hewlett-Packard (HP) Co. and BlueArc Corp. use virtualization to enable what amounts to thin provisioning in other forms. It's storage virtualization combined with detailed knowledge of the organization's data usage patterns that make thin provisioning effective.