A website's architecture has greater effect on effective capacity than any server or network tuning. Design determines both how efficiently you consume capacity and how easily you can add or subtract it as need arises. Keeping the architecture **divisible and segmented** is what lets you characterize load per component — the prerequisite for finding each component's red-line ceiling. Design also bounds your scaling options. Adding read-only replica databases scales reads, but only up to a point: beyond it the master's write volume outpaces what the replicas can sustain, no matter how many you add (diminishing returns Flickr hit directly). Sites then federate across multiple masters, split into clusters, or cache — architecture, not tuning, decides which paths stay open. --- *Source: [[The Art of Capacity Planning]] (John Allspaw, O'Reilly 2008) — Ch 1 — Goals, Issues, and Processes in Capacity Planning*