The Burton Group weighed in today with a post on thin provisioning.  While discussing both pros and cons, they tend to veer towards the dark side - and that's fine with me.  I think its important for people to understand the risks of things before they get themselves in trouble - like using credit cards or driving a car.  The thing that bugged me was that they didn't mention EqualLogic in their analysis.  I'd gladly put our algorithms and implementation up against anybody else's, and here is why:

Burton Group says array limitations are a problem.  Not with our distributed, page-based, frameless architecture.  When you run out of resources with an EqualLogic iSCSI storage system, put another one in the rack and let them discover each other (happens automatically) and form a group. If you ever want to upgrade the disk drives in one of our products you can do that too - without experiencing a lick of downtime. 

Burton Group says garbage collection - (or the lack of it, to be precise) is a problem. It certainly can be, and is probably the most important thing to keep in mind as an administrator of a thinly provisioned volume or system.  That said, there are lots of system and application scenarios where this is far less of a problem than chronic data growth.  In other words, the growth curve far outpaces the garbage that is being created.  Unstoppable capacity growth scenarios pay the highest dividends in terms of cost savings, capacity planning advantages and provisioning simplicity.

Burton Group says rogue processes that dump lots of random data into a volume can wreak havoc if it is thinly provisioned.  Yes they can, but this is not an immediate threat if you follow common best practices for limiting, isolating and fencing mission critical data from other users and processes in the SAN.  For instance, you probably don't want an Oracle or SQL Server application volume crushed because a renegade user that starts copying video files in the middle of the night.  So give your database applications and the file servers independent volumes. This is one of the reasons you want thin provisioning implemented at the volume level, as EqualLogic has. Not only that, but compared to Fibre Channel - supposedly a more "enterprise ready technology" , iSCSI offers far better authentication and security technologies to prevent inadvertent access to storage volumes.

Burton Group says degraded performance can be a problem.  Maybe, but not necessarily.  They claim that thin provisioning equates with fewer disk spindles; but that is not necessarily true either.  There is no reason the number of spindles has to decrease with a thinly provisioned volume. It's true that block mapping takes some small amount of time, but this is like saying that RAID mappings slow down I/O processes.  The impact is (or should be) minimal. FWIW, our products use all spindles for all thinly provisioned volumes.  If you run out of capacity in one of our systems (or if you want to improve performance) you can add another system to the group and double the number of spindles and network ports servicing your application.  EqualLogic's frameless architecture has linear performance scaling.

Burton Group also asks about alerts and performance throttling when thin-volume capacity hits certain threshold. Yes, yes, yes.  We have those scenarios covered too - alerts, thresholds and performance throttling are all in the product. 

Most importantly, we let customers convert between thin and normal volumes as they wish.  If there is any reason to turn off thin provisioning, they can.  One of the reasons they might want to do that is for applications that have matured and have seen their growth stabilize.  There is no need to worry about the lack of garbage collection if you don't have to. Turn off thin provisioning and manage it as a normal volume.