Have a bunch of MS Dynamics GP databases. The tables typically are substantially de-normalised (very "wide" tables of partially redundant data), have long fixed length character columns, and large numbers of decimal columns containing zeros. The databases and all their contained tables (some 1000 tables in a typical GP database) compress excellently using row compression; we haven't implemented page compression. In the large number of tables it appeared difficult to select appropriate candidates; it seemed easier to row compress all (except tables with a very low number of rows) and uncompress the exceptions if and when we found them. Have found no noticeably poor candidates as yet. In fairness to MS the table design is probably a legacy from the earlier multi-database support

In terms of CPU usage I am assuming that SELECTs with a high number of logical reads (in the hundereds of thousands of rows) should show high CPU when compressed. The effect does not appear noticeable, and the server in any case currently has horsepower to spare. Any effect is dwarfed by other CPU effects we may have from custom code

Since we are currently blessed with both sufficient RAM to contain the entire set of databases (nearly zero read IO from the smaller databases), and more recently some really fast write IO, our performance experience is a trifle difficult to compare

Tony T