Nice examination of the “Top 5″ sections. I have now included links to the pages in the book – the next step is to see if the “Top 5″ sections show what the author claims.

The numbers don’t appear to add up correctly in several of the examples and several number recur in the first eleven:

The db file sequential/scattered read events appear in 1, 5, 9, 10 and 11 with very similar/the same numbers for either waits or time, but differ slightly on the other columns each time.

In 4, the times for sequential and scattered read are very similar, yet have vastly different % of elapsed (29 vs 9). In 5, the events aren’t ordered by time, and the % elapsed don’t correspond to the times taken (the top event is over 80% of elapsed time, despite consuming less than half the time of the third).

The percentage of elapsed times add up to over 200% in seven; this seems unusual to me, though I may not be understanding something which is causing this.

The first eleven are also inconsistent with the display of the time column – sometime it’s shown and sometimes it isn’t…

I’d certainly be more than a little worried about a system (excerpt 1) that on average was taking 2.75 seconds to do a single block read. Perhaps unlike the imps in the wonderful iconograph these data imps have gone on strike. It’s possible of course that these figures are blown by bugs in the data – sometimes oracle has been known to print incorrect values in extended sql trace ela values, possibly some versions of statspack/awr have encountered similar issues.

The second extract has a value in the “Waits” column against CPU Time which should not be there and single block reads that take on average one second per read.