Arthur Fuller
fuller.artful at gmail.com
Sat Jan 26 13:46:41 CST 2013
The first question I'd ask, and doubtless you did, is, "How frequently will the old data ever be needed?" And the followup, "How far back?" Presumably the answer is a gradient, something like "the past five years, once a month; the previous five years before that, once a quarter or year; and the rest once in a blue moon." My reasoning is this. I'd prefer (SQL licensing and version permitting) to partition the tables across several hard disks, leaving only the past 5 years or so immediately accessible, and hitting the remote hard disks and partitions only occasionally. This would sidestep the need for any additional code or restore-procedures, etc. IIRC, this capability is available only in the more expensive licensing agreements and versions, which is why I asked about those. Another take on this, with similar perspective, is based on the simple notion that there is a "contemporary" window, beyond which data will never change. That suggests a possible move of all that old data into an Analysis Services database, which would be de-normalized and with all its counts and sums all available in pre-calculated form. That would require some serious space, but on the other hand, what in this business is cheaper than disk space? Arthur