This content has been marked as final. Show 3 replies
This is not really an Exadata-specific question, but more a function of workload. That having been said, I've this come up in an Exadata context when doing high-concurrency data changes, particularly unsynchronized RAT replays. The issue presented itself as a high number of buffer busy waits for block 2 (the LMT space management header block) of the single large tempfile. Createing a new temporary tablespace with many, smaller tempfiles resolved the issue.
MOS note 465840.1 also has some useful information.
I've put together a blog post with more details about my experiences:
Suggest using temp tablespace groups for best results... http://docs.oracle.com/cd/B28359_01/server.111/b28310/tspaces002.htm#i1008609