Matt Guthrie (6/6/2011)
Am I the only one who noticed that the write latency for TEMPDB was measured in *seconds*, not ms?
I would think 3.5 sec is a lot more worrisome than 3.5ms. Just sayin'.
You must be, as I missed this, and so did the technical reviewer. Good catch! Next time I will pay more attention to the scale, instead of making assumptions, which is apparently what I did when I looked at the graph.
This definitely needs more in-depth analysis to see what is really going on during this maintenance window within tempdb. I have already planned to do a specific follow-up article on tempdb, and I will add this to my list to investigate.