Indianrock (8/23/2013)

---

I suspect the daytime average write wait on tempdb is very low, but the average is driven very high by doing update statistics on tables with over 150 million records at an 80% sample rate.

1) You should never see wait times like 1.2 seconds on a reasonably tuned IO system. There is no reason to accept it.

2) As I recall FULLSCAN gives better performance than high sample rates. I don't remember exactly where the tipping point is, but I think it may be around 50%. I think Linchi Shea has a blog post on this.

I hold that as unlikely. Particularly when it is clear that you I/O subsystem needs an overhaul. You SAN guys need to review their configuration.