Commitc2df45a

committed

pgbench: Correct for bias in --rate schedule generation.

Previous code gave a mean delay 0.44% below target. This change alsohas the effect of increasing the maximum possible delay.Fabien COELHO

1 parent77ae7f7 commitc2df45aCopy full SHA for c2df45a

File tree

-3

lines changed

-3

lines changed

Lines changed: 7 additions & 3 deletions

Original file line number	Diff line number	Diff line change
`@@ -929,13 +929,17 @@ doCustom(TState thread, CState st, instr_time conn_time, FILE logfile, AggVa`
`929`	`929`	`* that the series of delays will approximate a Poisson distribution`
`930`	`930`	`* centered on the throttle_delay time.`
`931`	`931`	`*`
`932`		`- * 1000 implies a 6.9 (-log(1/1000)) to 0.0 (log 1.0) delay multiplier.`
	`932`	`+ * 10000 implies a 9.2 (-log(1/10000)) to 0.0 (log 1) delay multiplier,`
	`933`	`+ * and results in a 0.055 % target underestimation bias:`
	`934`	`+ *`
	`935`	`+ * SELECT 1.0/AVG(-LN(i/10000.0)) FROM generate_series(1,10000) AS i;`
	`936`	`+ * = 1.000552717032611116335474`
`933`	`937`	`*`
`934`	`938`	`* If transactions are too slow or a given wait is shorter than`
`935`	`939`	`* a transaction, the next transaction will start right away.`
`936`	`940`	`*/`
`937`		`-int64wait= (int64)`
`938`		`-throttle_delay*-log(getrand(thread,1,1000)/1000.0);`
	`941`	`+int64wait= (int64) (throttle_delay*`
	`942`	`+1.00055271703*-log(getrand(thread,1,10000)/10000.0));`
`939`	`943`
`940`	`944`	`thread->throttle_trigger+=wait;`
`941`	`945`

Comments

(0)