I suspect this must have been early in 9.2 or 10.2 – I’m fairly sure I remember seeing cases where the mreadtim was used in the arithmetic even when it was smaller than the sreadtim and the whole mreadtim / sreadtim adjustment should have been ignored.

We had a situation at a customer site where MREADTIM < SREADTIM. We then experimented with those system stats and table stats from the production DB in a test database. We never got the FTS back so I figured that the system stats must have had this effect. (I also found a comment with similar content somewhere.)

I think system stats don’t get mentioned much because they aren’t automatic in 9i, and the defaults in 10g are generally adequate so few people notice that they are there.

Certainly, though, there were some cases where odd things happened in the early days, in some versions, when people tried to use them in 9i and got odd values – so any trouble-shooting job should include a quick check on current values.

While we’re at it: I am under the impression that system stats are somewhat underestimated. One can read frequently about table and schema stats but it is rarely suggested in public forums to check system stats although they can have a dramatic impact on execution plans (namely if MREADTIM < SREADTIM). Does anybody else have that impression of underestimation?

Greetings