Annotation of embedaddon/sqlite3/tool/mkspeedsql.tcl, revision 1.1
1.1 ! misho 1: # 2008 October 9
! 2: #
! 3: # The author disclaims copyright to this source code. In place of
! 4: # a legal notice, here is a blessing:
! 5: #
! 6: # May you do good and not evil.
! 7: # May you find forgiveness for yourself and forgive others.
! 8: # May you share freely, never taking more than you give.
! 9: #
! 10: #*************************************************************************
! 11: # This file generates SQL text used for performance testing.
! 12: #
! 13: # $Id: mkspeedsql.tcl,v 1.1 2008/10/09 17:57:34 drh Exp $
! 14: #
! 15:
! 16: # Set a uniform random seed
! 17: expr srand(0)
! 18:
! 19: # The number_name procedure below converts its argment (an integer)
! 20: # into a string which is the English-language name for that number.
! 21: #
! 22: # Example:
! 23: #
! 24: # puts [number_name 123] -> "one hundred twenty three"
! 25: #
! 26: set ones {zero one two three four five six seven eight nine
! 27: ten eleven twelve thirteen fourteen fifteen sixteen seventeen
! 28: eighteen nineteen}
! 29: set tens {{} ten twenty thirty forty fifty sixty seventy eighty ninety}
! 30: proc number_name {n} {
! 31: if {$n>=1000} {
! 32: set txt "[number_name [expr {$n/1000}]] thousand"
! 33: set n [expr {$n%1000}]
! 34: } else {
! 35: set txt {}
! 36: }
! 37: if {$n>=100} {
! 38: append txt " [lindex $::ones [expr {$n/100}]] hundred"
! 39: set n [expr {$n%100}]
! 40: }
! 41: if {$n>=20} {
! 42: append txt " [lindex $::tens [expr {$n/10}]]"
! 43: set n [expr {$n%10}]
! 44: }
! 45: if {$n>0} {
! 46: append txt " [lindex $::ones $n]"
! 47: }
! 48: set txt [string trim $txt]
! 49: if {$txt==""} {set txt zero}
! 50: return $txt
! 51: }
! 52:
! 53: # Create a database schema.
! 54: #
! 55: puts {
! 56: PRAGMA page_size=1024;
! 57: PRAGMA cache_size=8192;
! 58: PRAGMA locking_mode=EXCLUSIVE;
! 59: CREATE TABLE t1(a INTEGER, b INTEGER, c TEXT);
! 60: CREATE TABLE t2(a INTEGER, b INTEGER, c TEXT);
! 61: CREATE INDEX i2a ON t2(a);
! 62: CREATE INDEX i2b ON t2(b);
! 63: SELECT name FROM sqlite_master ORDER BY 1;
! 64: }
! 65:
! 66:
! 67: # 50000 INSERTs on an unindexed table
! 68: #
! 69: set t1c_list {}
! 70: puts {BEGIN;}
! 71: for {set i 1} {$i<=50000} {incr i} {
! 72: set r [expr {int(rand()*500000)}]
! 73: set x [number_name $r]
! 74: lappend t1c_list $x
! 75: puts "INSERT INTO t1 VALUES($i,$r,'$x');"
! 76: }
! 77: puts {COMMIT;}
! 78:
! 79: # 50000 INSERTs on an indexed table
! 80: #
! 81: puts {BEGIN;}
! 82: for {set i 1} {$i<=50000} {incr i} {
! 83: set r [expr {int(rand()*500000)}]
! 84: puts "INSERT INTO t2 VALUES($i,$r,'[number_name $r]');"
! 85: }
! 86: puts {COMMIT;}
! 87:
! 88:
! 89: # 50 SELECTs on an integer comparison. There is no index so
! 90: # a full table scan is required.
! 91: #
! 92: for {set i 0} {$i<50} {incr i} {
! 93: set lwr [expr {$i*100}]
! 94: set upr [expr {($i+10)*100}]
! 95: puts "SELECT count(*), avg(b) FROM t1 WHERE b>=$lwr AND b<$upr;"
! 96: }
! 97:
! 98: # 50 SELECTs on an LIKE comparison. There is no index so a full
! 99: # table scan is required.
! 100: #
! 101: for {set i 0} {$i<50} {incr i} {
! 102: puts "SELECT count(*), avg(b) FROM t1 WHERE c LIKE '%[number_name $i]%';"
! 103: }
! 104:
! 105: # Create indices
! 106: #
! 107: puts {BEGIN;}
! 108: puts {
! 109: CREATE INDEX i1a ON t1(a);
! 110: CREATE INDEX i1b ON t1(b);
! 111: CREATE INDEX i1c ON t1(c);
! 112: }
! 113: puts {COMMIT;}
! 114:
! 115: # 5000 SELECTs on an integer comparison where the integer is
! 116: # indexed.
! 117: #
! 118: set sql {}
! 119: for {set i 0} {$i<5000} {incr i} {
! 120: set lwr [expr {$i*100}]
! 121: set upr [expr {($i+10)*100}]
! 122: puts "SELECT count(*), avg(b) FROM t1 WHERE b>=$lwr AND b<$upr;"
! 123: }
! 124:
! 125: # 100000 random SELECTs against rowid.
! 126: #
! 127: for {set i 1} {$i<=100000} {incr i} {
! 128: set id [expr {int(rand()*50000)+1}]
! 129: puts "SELECT c FROM t1 WHERE rowid=$id;"
! 130: }
! 131:
! 132: # 100000 random SELECTs against a unique indexed column.
! 133: #
! 134: for {set i 1} {$i<=100000} {incr i} {
! 135: set id [expr {int(rand()*50000)+1}]
! 136: puts "SELECT c FROM t1 WHERE a=$id;"
! 137: }
! 138:
! 139: # 50000 random SELECTs against an indexed column text column
! 140: #
! 141: set nt1c [llength $t1c_list]
! 142: for {set i 0} {$i<50000} {incr i} {
! 143: set r [expr {int(rand()*$nt1c)}]
! 144: set c [lindex $t1c_list $i]
! 145: puts "SELECT c FROM t1 WHERE c='$c';"
! 146: }
! 147:
! 148:
! 149: # Vacuum
! 150: puts {VACUUM;}
! 151:
! 152: # 5000 updates of ranges where the field being compared is indexed.
! 153: #
! 154: puts {BEGIN;}
! 155: for {set i 0} {$i<5000} {incr i} {
! 156: set lwr [expr {$i*2}]
! 157: set upr [expr {($i+1)*2}]
! 158: puts "UPDATE t1 SET b=b*2 WHERE a>=$lwr AND a<$upr;"
! 159: }
! 160: puts {COMMIT;}
! 161:
! 162: # 50000 single-row updates. An index is used to find the row quickly.
! 163: #
! 164: puts {BEGIN;}
! 165: for {set i 0} {$i<50000} {incr i} {
! 166: set r [expr {int(rand()*500000)}]
! 167: puts "UPDATE t1 SET b=$r WHERE a=$i;"
! 168: }
! 169: puts {COMMIT;}
! 170:
! 171: # 1 big text update that touches every row in the table.
! 172: #
! 173: puts {
! 174: UPDATE t1 SET c=a;
! 175: }
! 176:
! 177: # Many individual text updates. Each row in the table is
! 178: # touched through an index.
! 179: #
! 180: puts {BEGIN;}
! 181: for {set i 1} {$i<=50000} {incr i} {
! 182: set r [expr {int(rand()*500000)}]
! 183: puts "UPDATE t1 SET c='[number_name $r]' WHERE a=$i;"
! 184: }
! 185: puts {COMMIT;}
! 186:
! 187: # Delete all content in a table.
! 188: #
! 189: puts {DELETE FROM t1;}
! 190:
! 191: # Copy one table into another
! 192: #
! 193: puts {INSERT INTO t1 SELECT * FROM t2;}
! 194:
! 195: # Delete all content in a table, one row at a time.
! 196: #
! 197: puts {DELETE FROM t1 WHERE 1;}
! 198:
! 199: # Refill the table yet again
! 200: #
! 201: puts {INSERT INTO t1 SELECT * FROM t2;}
! 202:
! 203: # Drop the table and recreate it without its indices.
! 204: #
! 205: puts {BEGIN;}
! 206: puts {
! 207: DROP TABLE t1;
! 208: CREATE TABLE t1(a INTEGER, b INTEGER, c TEXT);
! 209: }
! 210: puts {COMMIT;}
! 211:
! 212: # Refill the table yet again. This copy should be faster because
! 213: # there are no indices to deal with.
! 214: #
! 215: puts {INSERT INTO t1 SELECT * FROM t2;}
! 216:
! 217: # Select 20000 rows from the table at random.
! 218: #
! 219: puts {
! 220: SELECT rowid FROM t1 ORDER BY random() LIMIT 20000;
! 221: }
! 222:
! 223: # Delete 20000 random rows from the table.
! 224: #
! 225: puts {
! 226: DELETE FROM t1 WHERE rowid IN
! 227: (SELECT rowid FROM t1 ORDER BY random() LIMIT 20000);
! 228: }
! 229: puts {SELECT count(*) FROM t1;}
! 230:
! 231: # Delete 20000 more rows at random from the table.
! 232: #
! 233: puts {
! 234: DELETE FROM t1 WHERE rowid IN
! 235: (SELECT rowid FROM t1 ORDER BY random() LIMIT 20000);
! 236: }
! 237: puts {SELECT count(*) FROM t1;}
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>