re: did you perform these deletes on the same dataserver with the same configurations
I was referring to the original comparison of partitioned vs non-partitoned table, ie, were these performed on the same dataserver with the same configurations? [In particular I was looking to see if perhaps statement cache was enabled on one dataserver but not the other ... a moot point if both tables are on the same dataserver and the deletes were run under the same config settings.]
--------------------
- logical IOs for the individual queries are identical, so the individual queries themselves don't appear to be to blame for the larger total logical IO count for the part table test
- the same logical IOs at the individual query level would also tend to rule out an issue with one table and/or index being more fragmented than the other table/index ("sp_spaceused <table>,1" would probably confirm similar sizes for the table and indexes of interest)
- cpu for the individual queries has some spikes for the non-part table, but the total cpu for the non-part is lower, so again the individual queries don't appear to be to blame for the larger total cpu count for the part table test
- are you by any chance running the non-part test with more options enabled (eg, 'set statistics io on' enabled for non-part test but disabled for the part test)? if you are running with any settings enabled (eg, 'set statistics', 'set showplan', etc) could you try running both tests with no extra settings enabled? [some settings ... like 'set statistics' and 'set showplan' can add considerably to your total cpu/logicalIO counts]