performance – Page 50 – Türker Yıldırım

SQL_CALC_FOUND_ROWS is usually slower than a second COUNT

NoteSeptember 25, 20151 min read

SQL_CALC_FOUND_ROWS promises the total row count for free alongside a LIMITed page, in one query. The catch is that it forces MySQL to evaluate every matching row rather…

A covering index answers the query without touching rows

NoteSeptember 24, 20151 min read

An index lookup normally finds a primary key and then goes to the table to fetch the columns you asked for — a second read per row. If…

microtime benchmarking measures your loop as well

NoteSeptember 8, 20151 min read

Wrapping a call in two microtime(true) readings looks like a measurement and is mostly noise: one iteration, no warm-up, and the timing includes whatever the loop and the…

PHP-FPM pm=dynamic numbers come from RAM, not from taste

NoteSeptember 3, 20151 min read

The default pool ships with pm.max_children = 50, which on a small server is not a limit so much as an invitation to swap. The number should come…

dump-autoload -o is not optional in production

NoteSeptember 3, 20151 min read

PSR-4 resolves a class by converting the namespace to a path and checking the filesystem. That is one or more stat calls per class, on every request, for…

Generators keep a million-row export inside memory

NoteAugust 18, 20152 min read

A method that builds an array and returns it holds every row in memory at once. For an export of a few hundred rows that is invisible; for…

ON DUPLICATE KEY UPDATE is one round trip, not two

NoteJune 30, 20151 min read

The usual upsert is a SELECT to see whether the row exists and then an INSERT or an UPDATE. That is two round trips, and between them another…

A Redis cache layer that does not go stale

Deep diveJune 10, 20159 min read

The cache made it fast and made it wrong. Key design, invalidation on write rather than on a timer, and why KEYS is the command that takes the…

will-change helps until it exhausts compositor memory

NoteMay 5, 20151 min read

will-change tells the browser to promote an element to its own compositor layer before the animation starts, which removes the stutter on the first frame. The advice usually…

Redis pipelining removes the round trips, not the work

NoteApril 30, 20151 min read

A hundred Redis commands issued one at a time cost a hundred network round trips. On the same host that is around a millisecond total; across an availability…