I am interested in any comments from users of Teradata regarding Teradata v13 Virtual Storage. Has anyone successfully implemented it in a live environment? an Active EDW? I am interested in real life experience rather than test setups or strictly controlled POC environments (Im a committed cynic!)
I have worked on a DB2 system about 6-7 years ago that employed similar technology via an EMC SAN device, this identified the different temperatures of disk and migrated it automatically as it saw fit. Whilst on paper it worked ok, our Warehouse at the time was more "active/operational" and as such the system suffered from inconsistant response times and a poor user experience.
If you implemented it, did you install from ground up, or upgrade an existing Configuration? If so How did you add the low cost slow disks, into the envrionment?
In the dev excahneg article http://developer.teradata.com/database/articles/teradata-virtual-storage-the-new-way-to-manage-multi-temperature-data the author claims that its not appropriate to implement in an Active/operation Datawarehouse. I found this comment at odds to what our management here are expecting.
Another couple of points on the virtualised storage.
It seems to move away from the "Shared nothing" type environment, with Amps sharing pools of disks. I would expect this should reduce performance? Also queries may read data from various types of disk, hot, medium, and cold, thus the query would run to the speed of the slowest part.
Data could be identified as cold when queried via one dimension, but required to be hot when queried by another.
If the data is moved on the basis of if its determined as hot, the data that is accessed by frequent interactive type queries (such as web based/or single amp queries) may skew the algorithm to mark those as hot, but the business benefit from the warehouse may come from desision support type queries which frequently mine data. Thus you may get one type of analytic user quering data every 5 minutes being relegated to cold storage, as ad hoc views from a non-business critical application are being queried at higher rates?