I've got a relatively simple query -
 select s.EDIBrchCd, S.ProdCd, B.Name
 from ScrubError s, Branch b
 where b.Hqid='ABC1234'
 and s.EDIBrchCd = b.BrchId

(A headquarter contains many branches - I want everything in
the  
ScrubError table for a single headquarter id....)

The showplan shows an access to the Branch table first, using a

non-clustered index on Hqid (there is only one qualifying
resulting row  
in this case,) then does a nested-iteration to ScrubError using
another  
non-clustered index on EDIBrchCd (there are NO rows in
ScrubError from  
the resulting branch-id) - the query plan seems VERY  
REASONABLE...however,

The statistics io show 42837 logical reads to ScrubError (!!!!
-  
coincidentally the number of total rows in ScrubError), and 3
to Branch  
(this part seems ok.)

Can somebody explain the 42837 number to me??  Seems like with
the  
non-clustered index, it should be doing about 2 or 3 reads.  
It's really  

difficult to optimize when the 'optimizer' is giving strange
results.

Thanks,