6 It is common for IO reads to complete in less than 1 ms. When this occurs, you know non-Oracle caching is involved. No physically spinning device can respond within 1 ms. Obviously, the block(s) requested does not reside in Oracle's buffer cache; otherwise, no wait event would have been posted, but you do know the block resides in some other cache. It could be the file system buffer cache, or perhaps that very nice cache you purchased from your IO subsystem vendor.

7 Another stellar network command is traceroute. "Hangs" in SQL*Net can commonly be discovered by running a few of these.

This chapter is different. Sure, the previous chapters have focused on Oracle performance diagnosis, but in this chapter, I pull everything together, filling in any missing pieces and introducing some additional diagnosis topics. This chapter will also complete the conceptual framework of performance diagnosis coverage, paving the way for the next three chapters on Oracle internals.