We have a data warehouse which serves up our BI system. Standard deal: we perform overnight loads, then all user-generated queries against the system are all read-based. No new data is written during the day.
The main database (we'll call it DW) is on our SAN in a RAID10 configuration. We have tempdb on its own separate RAID10 LUN. Generally, system performance is solid. However, on Monday mornings in particular, the system gets utterly hammered. This is increasingly becoming a problem. I did some profiling this morning as the system was hammered, and wanted to share my results.
When users are all accessing DW, I'm seeing PAGEIOLATCH_SH waits like crazy.
I see the CPU staying around 10%, with occasional jumps up to ~80%, but mostly seems to be waiting on the disks. Current disk queue length is spiking constantly for hours straight while...