Dell Compellent's auto tiering feature is called Data Progression. It works by progressing infrequently accessed pages (2MB by default) down a tier and frequently access pages up a tier (in theory at least). The tiers available for any given volume are controller with a policy called a Storage Profile. Something to keep in mind is that a tier is commonly referenced as SSD, 15k, 10k or 7.2k, but in reality what the Compellent system views as a tier is far more granular than this. A tier in Storage Center (Compellent's operating system) is truly defined as the combination of a disk class (SSD, 15k, 10k, 7.2k) and RAID type (10, 5, or 6) and lastly on spinning disks by track (fast track vs standard). The last of those tiering option is a an additional license called Fast Track and it basically segments the spinning disk platters so the outer 20% of the disk tracks are used as fast tracks and the remaining 80% is standard.
So, data doesn't just move down from one class of disk to another, it also moves from outside of the disk to the inside of the disk while also changing the raid level used to protect the data. What this ultimately buys you is your most frequently accessed data (and new writes if setup correctly) will always be on your highest class of disk and potentially in the outer tracks to short stroke the spinning drives. The movement from raid 10 to raid 5/6 is the gain higher capacity utilization and typically is the very first move that will take place for data.
Now, I know some things have changed with Compellent since I was working there last in November 2012, but even if the schedule has been adjusted to occur more frequently than once every 24 hours, you still have a scheduled task (called Data Progression) to analyze and then move the data and to your question, yes it can be problematic in environments where data might sit idle for a period of time before being accessed again. But data will never move unless you take the minimum 1 Replay (snapshot) per Data Progression cycle. Back to the Storage Profile I mentioned earlier, it has check boxes where you can check what tiers (disk class and raid type only) you want the system to allow volumes to use for writable data and a separate column of checkboxes for how you want replay data (snapshots) to be stored. So, the idea is you take a snapshot at least once per data progression cycle (preferably just before data progression would start and your data would stay on the same class of disk, but would be moved read only from RAID 10 to RAID 5 or 6. Your new writes would always come in at the highest available tier for the given storage profile, which is most commonly raid 10 on the fastest disk type.
In theory, reading from RAID 5 or 6 should be nearly identical in performance to RAID 10, but your mileage will fluctuate. Again, in theory, if your lowest class of disks aren't busy trying to handle writes they should be very quick to service read requests for data that had progressed down to them and as the developers kick off new builds which write out new files those writes will go directly to the highest tier and thus begins a new page of data to start its journey through the tiers.
Hopefully you've found this to be a little bit helpful. If you have any other questions please feel free to ask.
I couldn't let this go without stating, you really might want to consider another option for your storage needs. Given your scenario, I have a feeling a system that can provide you with in-line data compression and in-line de-duplication might really help you provide a far superior level of response to your developers while reducing your storage footprint. I left Dell almost 2 years after coming over from the Compellent acquisition because the product really stopped moving anywhere. I'm biased to the company I work for now, Tegile, but any of the flash or hybrid players with data reduction deserve at least a mention and some investigation for use in your environment.
Best of luck in your search for your next storage solution
Sincerely,
Aaron Gipson