Preparing for growth.
WHAT TO PROJECT
Data volume Query volume Concurrency Working set against memory
WHY THE WORKING SET MATTERS MOST
Performance falls sharply when it no longer fits in memory.
HOW TO ESTIMATE IT
Measure which tables and indexes are actually read.
WHAT TO MEASURE OVER TIME
Table sizes Query rate Response times at percentiles Buffer pool hit rate
WHY PERCENTILES
Averages conceal the slow queries users notice.
WHAT TO PROJECT FROM
Growth over several months, not one.
WHAT TO PLAN BEFORE
Memory exhaustion Disk exhaustion Connection limits
WHAT TO CONSIDER AT EACH THRESHOLD
Indexing and query improvements first Archiving second More memory third Replicas or splitting last
WHY THAT ORDER
Each earlier step is cheaper and frequently sufficient.
WHAT TO TEST BEFORE COMMITTING TO HARDWARE
Whether the workload is actually constrained by what you intend to add.
WHAT TO AVOID
Scaling hardware to conceal an unindexed query.
WHY
It converts a fixable problem into a permanent cost.
WHAT TO REVIEW QUARTERLY
Growth against projection.
