SQ32@LIVERPOOL.AC.UK (10/26/90)
Hello, I am trying to establish which of the storage structures is best for my data, which has some characteristics in its evolution, and I am using a benchmark of of queries to help me do that. The symbolic constants one can use in INGRES (BTW, I am using INGRES 5.0/06 on HP9000/850 [release 7.0]) are: 1. _et_sec : gives elapsed time. 2. _cpu_ms : gives cpu time. 3. _dio_cnt : gives direct I/O requests. 4. _pfault_cnt : gives page faults. 1 and 2 do not give accurate measurements (depend on the system workload), even though 2 is better in a sense that the numbers I get for the same query at different times are not that much different (they are though!) compared to the ones I get when using 1. However, I have been told that to get solid results one has to use 4. The problem is 3 and 4 give me 0 no matter how large my relations are (and they are: the average is 3000 tuples whose length is 124 bytes each). Does anybody know why this is happening? How to deal with it? Or how to get round it? Any comments or suggestions would be fully appreciated. Thanks in advance.