I read a couple interesting posts related to the importance of using a holistic system view for DB tuning, rather than using a single criterion such as wait stats.
https://www.brentozar.com/archive/2020/01/why-database-monitoring-tools-are-so-hard-to-interpret/
https://sqlperformance.com/2020/01/sql-performance/why-waits-alone-are-not-enough
The comprehensive hourly-load view on Ozar’s post in particular implicitly highlighted a key difference in database tuning for on-premise servers versus cloud-based servers. That is, on-prem CPU cycles are (arguably) free. Cloud-based CPU cycles are not. This means the organization pays for excess CPU cycles for cloud-based systems. This completely changes the paradigm of “good enough” tuning.
If you have an unexplained case of any condition, your spe lowest prices cialist will look forward to check your prostate gland. It is a useful cheap viagra in australia medication for pulmonary arterial hypertension and benign prostate hypertrophy. Some of viagra shop usa them are: Online drivers ed programs available to them to fulfill their state required education obligations. Kamagra and getting viagra in canadae both mainstream approaches to manage erectile dysfunction and testicular pain.Instead of stopping a tuning project when a query does not have a practical impact upon system performance (because cost is not a concern), a cloud-based database can ALWAYS benefit from queries that drive fewer CPU cycles (and also storage costs, etc.) This requires additional decisions during tuning, considering ROI (return on investment) of the costs of the tuning activity. Given a known cost per CPU cycle, and the expected delta in the number of CPU cycles, and the traditionally-critical impact on performance, now the cost/benefit analysis can help find the stopping point for query tuning.
Taking the cost factor into consideration is also important in choosing which queries to tune. And the frequency of regularly-scheduled queries. If a periodic query is costing the organization $10M in annual CPU cycles, it’s tuning will generally be vastly more important than tweaking an online transaction query from five seconds down to three, if that tweak saves the organization $200K annually.
I appreciate Ozar’s and Gonzalez’ posts that led me to this understanding.