Todays’ SQL Server TIPS was on LazyWriter, and recommended adding spindles to improve I/O performance. That seems simplified to the point of uselessness. What about systems that can’t integrate ad-hoc additions? How to re-balance the actual data across spindles? Are the performance hits from random or sequential, read or update, batch or interactive? What about the components of a “disk” subsytem besides number of spindles? Cache, spin latency, SCSI flavour, interface (wire or fibre), host adapter throughput? Database design (appropriate indexing), query design (using those indices), …
While I would not expect a daily TIP to cover all this in any depth (many volumes have been written trying to do that :), it should have a couple of sentences to remind everyone that blindly throwing hardware at a problem is rarely a good long term solution.
