Skip to content

H
h2database
切换分支/标签
查找文件
Normal viewHistoryPermalink
performance.html 19.7 KB
Newer Older
Thomas Mueller's avatar
--no commit message
35e27cd5
 
Thomas Mueller committed
1 2 
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
<!--
Thomas Mueller's avatar
--no commit message
483d9103
 
Thomas Mueller committed
3 
Copyright 2004-2007 H2 Group. Licensed under the H2 License, Version 1.0 (http://h2database.com/html/license.html).
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
4 5 
Initial Developer: H2 Group
-->
Thomas Mueller's avatar
--no commit message
35e27cd5
 
Thomas Mueller committed
6 7 
<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en">
<head><meta http-equiv="Content-Type" content="text/html;charset=utf-8" /><title>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
8 
Performance
Thomas Mueller's avatar
--no commit message
b492f851
 
Thomas Mueller committed
9 
</title><link rel="stylesheet" type="text/css" href="stylesheet.css" />
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
10 11 12 13 14 15 
<script type="text/javascript" src="navigation.js"></script>
</head><body onload="frameMe();">
<table class="content"><tr class="content"><td class="content"><div class="contentDiv">

<h1>Performance</h1>
<a href="#performance_comparison">
Thomas Mueller's avatar
--no commit message
b492f851
 
Thomas Mueller committed
16 
    Performance Comparison</a><br />
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
17 
<a href="#application_profiling">
Thomas Mueller's avatar
--no commit message
b492f851
 
Thomas Mueller committed
18 
    Application Profiling</a><br />
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
19 
<a href="#database_performance_tuning">
Thomas Mueller's avatar
--no commit message
b492f851
 
Thomas Mueller committed
20 
    Performance Tuning</a><br />
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
21
Thomas Mueller's avatar
--no commit message
b492f851
 
Thomas Mueller committed
22 
<br /><a name="performance_comparison"></a>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
23 
<h2>Performance Comparison</h2>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
24 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
25 
In most cases H2 is a lot faster than all other
Thomas Mueller's avatar
--no commit message
927afe2d
 
Thomas Mueller committed
26 27 
(open source and not open source) database engines. 
Please note this is mostly a single connection benchmark run on one computer.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
28 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
29 30 31 32 

<h3>Embedded</h3>
<table border="1" class="bar">
<tr><th>Test Case</th><th>Unit</th><th>H2</th><th>HSQLDB</th><th>Derby</th></tr>
Thomas Mueller's avatar
--no commit message
dfbbf263
 
Thomas Mueller committed
33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 
<tr><td>Simple: Init</td><td>ms</td><td>375</td><td>578</td><td>2797</td></tr>
<tr><td>Simple: Query (random)</td><td>ms</td><td>250</td><td>344</td><td>1563</td></tr>
<tr><td>Simple: Query (sequential)</td><td>ms</td><td>171</td><td>250</td><td>1469</td></tr>
<tr><td>Simple: Update (random)</td><td>ms</td><td>641</td><td>1609</td><td>19265</td></tr>
<tr><td>Simple: Delete (sequential)</td><td>ms</td><td>172</td><td>516</td><td>6797</td></tr>
<tr><td>Simple: Memory Usage</td><td>MB</td><td>14</td><td>12</td><td>12</td></tr>
<tr><td>BenchA: Init</td><td>ms</td><td>391</td><td>500</td><td>3750</td></tr>
<tr><td>BenchA: Transactions</td><td>ms</td><td>5468</td><td>2468</td><td>16250</td></tr>
<tr><td>BenchA: Memory Usage</td><td>MB</td><td>14</td><td>15</td><td>9</td></tr>
<tr><td>BenchB: Init</td><td>ms</td><td>1281</td><td>2391</td><td>14938</td></tr>
<tr><td>BenchB: Transactions</td><td>ms</td><td>2094</td><td>1140</td><td>3828</td></tr>
<tr><td>BenchB: Memory Usage</td><td>MB</td><td>16</td><td>11</td><td>9</td></tr>
<tr><td>BenchC: Init</td><td>ms</td><td>984</td><td>547</td><td>5250</td></tr>
<tr><td>BenchC: Transactions</td><td>ms</td><td>2860</td><td>58219</td><td>11204</td></tr>
<tr><td>BenchC: Memory Usage</td><td>MB</td><td>19</td><td>19</td><td>9</td></tr>
Thomas Mueller's avatar
--no commit message
b09fd536
 
Thomas Mueller committed
48 
<tr><td>Executed Statements</td><td>#</td><td>594255</td><td>594255</td><td>594255</td></tr>
Thomas Mueller's avatar
--no commit message
dfbbf263
 
Thomas Mueller committed
49 50 
<tr><td>Total Time</td><td>ms</td><td>14687</td><td>68562</td><td>87111</td></tr>
<tr><td>Statement per Second</td><td>#</td><td>40461</td><td>8667</td><td>6821</td></tr>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
51 52 53 54 55 
</table>

<h3>Client-Server</h3>
<table border="1" class="bar">
<tr><th>Test Case</th><th>Unit</th><th>H2</th><th>HSQLDB</th><th>Derby</th><th>PostgreSQL</th><th>MySQL</th></tr>
Thomas Mueller's avatar
--no commit message
dfbbf263
 
Thomas Mueller committed
56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 
<tr><td>Simple: Init</td><td>ms</td><td>3047</td><td>2547</td><td>6907</td><td>4234</td><td>3594</td></tr>
<tr><td>Simple: Query (random)</td><td>ms</td><td>3547</td><td>2641</td><td>8781</td><td>5375</td><td>3140</td></tr>
<tr><td>Simple: Query (sequential)</td><td>ms</td><td>3390</td><td>2531</td><td>8859</td><td>4906</td><td>3016</td></tr>
<tr><td>Simple: Update (random)</td><td>ms</td><td>3235</td><td>3531</td><td>22344</td><td>5828</td><td>5187</td></tr>
<tr><td>Simple: Delete (sequential)</td><td>ms</td><td>1421</td><td>1235</td><td>8219</td><td>2484</td><td>1829</td></tr>
<tr><td>Simple: Memory Usage</td><td>MB</td><td>15</td><td>10</td><td>15</td><td>0</td><td>0</td></tr>
<tr><td>BenchA: Init</td><td>ms</td><td>2687</td><td>2343</td><td>6000</td><td>4000</td><td>4000</td></tr>
<tr><td>BenchA: Transactions</td><td>ms</td><td>12938</td><td>9579</td><td>26610</td><td>16250</td><td>10782</td></tr>
<tr><td>BenchA: Memory Usage</td><td>MB</td><td>15</td><td>16</td><td>10</td><td>0</td><td>0</td></tr>
<tr><td>BenchB: Init</td><td>ms</td><td>9641</td><td>10094</td><td>28282</td><td>17468</td><td>11344</td></tr>
<tr><td>BenchB: Transactions</td><td>ms</td><td>3984</td><td>3312</td><td>6671</td><td>7797</td><td>3375</td></tr>
<tr><td>BenchB: Memory Usage</td><td>MB</td><td>16</td><td>13</td><td>8</td><td>0</td><td>0</td></tr>
<tr><td>BenchC: Init</td><td>ms</td><td>2031</td><td>1516</td><td>7391</td><td>2297</td><td>3406</td></tr>
<tr><td>BenchC: Transactions</td><td>ms</td><td>9750</td><td>58734</td><td>20937</td><td>11172</td><td>7469</td></tr>
<tr><td>BenchC: Memory Usage</td><td>MB</td><td>20</td><td>15</td><td>14</td><td>0</td><td>0</td></tr>
Thomas Mueller's avatar
--no commit message
b09fd536
 
Thomas Mueller committed
71 
<tr><td>Executed Statements</td><td>#</td><td>594255</td><td>594255</td><td>594255</td><td>594255</td><td>594255</td></tr>
Thomas Mueller's avatar
--no commit message
dfbbf263
 
Thomas Mueller committed
72 73 
<tr><td>Total Time</td><td>ms</td><td>55671</td><td>98063</td><td>151001</td><td>81811</td><td>57142</td></tr>
<tr><td>Statement per Second</td><td>#</td><td>10674</td><td>6059</td><td>3935</td><td>7263</td><td>10399</td></tr>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
74 75 76 77 78 
</table>

<h3>Benchmark Results and Comments</h3>

<h4>H2</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
79 
<p>
Thomas Mueller's avatar
--no commit message
dfbbf263
 
Thomas Mueller committed
80 
Version 1.0 (2007-09-15) was used for the test.
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
81 82 83 84 85 86 87 88 
For simpler operations, the performance of H2 is about the same as for HSQLDB.
For more complex queries, the query optimizer is very important.
However H2 is not very fast in every case, certain kind of queries may still be slow.
One situation where is H2 is slow is large result sets, because they are buffered to
disk if more than a certain number of records are returned.
The advantage of buffering is, there is no limit on the result set size.
The open/close time is almost fixed, because of the file locking protocol: The engine waits
20 ms after opening a database to ensure the database files are not opened by another process.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
89 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
90 91 

<h4>HSQLDB</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
92 
<p>
Thomas Mueller's avatar
--no commit message
dfbbf263
 
Thomas Mueller committed
93 
Version 1.8.0.8 was used for the test.
Thomas Mueller's avatar
--no commit message
9d0f0920
 
Thomas Mueller committed
94 95 
Cached tables are used in this test (hsqldb.default_table_type=cached),
and the write delay is 1 second (SET WRITE_DELAY 1).
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
96 97 98 99 100 101 102 103 104 105 106 
HSQLDB is fast when using simple operations.
HSQLDB is very slow in the last test (BenchC: Transactions), probably because is has a bad query optimizer.
One query where HSQLDB is slow is a two-table join:
<pre>
SELECT COUNT(DISTINCT S_I_ID) FROM ORDER_LINE, STOCK
WHERE OL_W_ID=? AND OL_D_ID=? AND OL_O_ID&lt;? AND OL_O_ID>=?
AND S_W_ID=? AND S_I_ID=OL_I_ID AND S_QUANTITY&lt;?
</pre>
The PolePosition benchmark also shows that the query optimizer does not do a very good job for some queries.
A disadvantage in HSQLDB is the slow startup / shutdown time (currently not listed) when using bigger databases.
The reason is, a backup of the database is created whenever the database is opened or closed.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
107 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
108 109 

<h4>Derby</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
110 
<p>
Thomas Mueller's avatar
--no commit message
dfbbf263
 
Thomas Mueller committed
111 
Version 10.3.1.4 was used for the test. Derby is clearly the slowest embedded database in this test.
Thomas Mueller's avatar
--no commit message
b09fd536
 
Thomas Mueller committed
112 113 
This seems to be a structural problem, because all operations are really slow. 
It will not be easy for the developers of Derby to improve the performance to a reasonable level.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
114 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
115 116 

<h4>PostgreSQL</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
117 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
118 119 120 121 
Version 8.1.4 was used for the test.
The following options where changed in postgresql.conf:
fsync = off, commit_delay = 1000.
PostgreSQL is run in server mode. It looks like the base performance is slower than
Thomas Mueller's avatar
--no commit message
9d0f0920
 
Thomas Mueller committed
122 
MySQL, the reason could be the network layer.
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
123 
The memory usage number is incorrect, because only the memory usage of the JDBC driver is measured.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
124 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
125 126 

<h4>MySQL</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
127 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
128 129 130 131 132 133 134 135 136 137 138 
Version 5.0.22 was used for the test.
MySQL was run with the InnoDB backend.
The setting innodb_flush_log_at_trx_commit
(found in the my.ini file) was set to 0. Otherwise (and by default), MySQL is really slow
(around 140 statements per second in this test) because it tries to flush the data to disk for each commit.
For small transactions (when autocommit is on) this is really slow.
But many use cases use small or relatively small transactions.
Too bad this setting is not listed in the configuration wizard,
and it always overwritten when using the wizard.
You need to change this setting manually in the file my.ini, and then restart the service.
The memory usage number is incorrect, because only the memory usage of the JDBC driver is measured.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
139 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
140 141 

<h4>Firebird</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
142 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
143 144 145 
Firebird 1.5 (default installation) was tested, but the results are not published currently.
It is possible to run the performance test with the Firebird database,
and any information on how to configure Firebird for higher performance are welcome.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
146 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
147 148 

<h4>Why Oracle / MS SQL Server / DB2 are Not Listed</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
149 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
150 151 152 153 
The license of these databases does not allow to publish benchmark results.
This doesn't mean that they are fast. They are in fact quite slow,
and need a lot of memory. But you will need to test this yourself.
SQLite was not tested because the JDBC driver doesn't support transactions.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
154 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
155 156 157 158 

<h3>About this Benchmark</h3>

<h4>Number of Connections</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
159 
<p>
Thomas Mueller's avatar
--no commit message
927afe2d
 
Thomas Mueller committed
160 161 
This is mostly a single-connection benchmark. 
BenchB uses multiple connections, the other tests one connection.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
162 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
163 164 

<h4>Real-World Tests</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
165 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
166 167 168 
Good benchmarks emulate real-world use cases. This benchmark includes 3 test cases:
A simple test case with one table and many small updates / deletes.
BenchA is similar to the TPC-A test, but single connection / single threaded (see also: www.tpc.org).
Thomas Mueller's avatar
--no commit message
927afe2d
 
Thomas Mueller committed
169 
BenchB is similar to the TPC-B test, using multiple connections (one thread per connection).
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
170 
BenchC is similar to the TPC-C test, but single connection / single threaded.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
171 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
172 173 

<h4>Comparing Embedded with Server Databases</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
174 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
175 176 177 178 
This is mainly a benchmark for embedded databases (where the application runs in the same
virtual machine than the database engine). However MySQL and PostgreSQL are not Java
databases and cannot be embedded into a Java application.
For the Java databases, both embedded and server modes are tested.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
179 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
180 181 

<h4>Test Platform</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
182 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
183 
This test is run on Windows XP with the virus scanner switched off.
Thomas Mueller's avatar
--no commit message
b09fd536
 
Thomas Mueller committed
184 
The VM used is Sun JDK 1.5.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
185 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
186 187 

<h4>Multiple Runs</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
188 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
189 190 191 192 
When a Java benchmark is run first, the code is not fully compiled and
therefore runs slower than when running multiple times. A benchmark
should always run the same test multiple times and ignore the first run(s).
This benchmark runs three times, the last run counts.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
193 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
194 195 

<h4>Memory Usage</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
196 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
197 198 
It is not enough to measure the time taken, the memory usage is important as well.
Performance can be improved in databases by using a bigger in-memory cache,
Thomas Mueller's avatar
--no commit message
bccf4e46
 
Thomas Mueller committed
199 
but there is only a limited amount of memory available on the system.
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
200 201 202 203 204 
HSQLDB tables are kept fully in memory by default, this benchmark
uses 'disk based' tables for all databases.
Unfortunately, it is not so easy to calculate the memory usage of PostgreSQL
and MySQL, because they run in a different process than the test. This benchmark currently
does not print memory usage of those databases.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
205 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
206 207 

<h4>Delayed Operations</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
208 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
209 210 211 
Some databases delay some operations (for example flushing the buffers)
until after the benchmark is run. This benchmark waits between
each database tested, and each database runs in a different process (sequentially).
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
212 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
213 214 

<h4>Transaction Commit / Durability</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
215 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
216 217 218 219 220 221 222 223 
Durability means transaction committed to the database will not be lost.
Some databases (for example MySQL) try to enforce this by default by calling fsync() to flush the buffers, but
most hard drives don't actually flush all data. Calling fsync() slows down transaction commit a lot,
but doesn't always make data durable. When comparing the results, it is important to
think about the effect. Many database suggest to 'batch' operations when possible.
This benchmark switches off autocommit when loading the data, and calls commit after each 1000
inserts. However many applications need 'short' transactions at runtime (a commit after each update).
This benchmark commits after each update / delete in the simple benchmark, and after each
Thomas Mueller's avatar
--no commit message
9d0f0920
 
Thomas Mueller committed
224 225 
business transaction in the other benchmarks. For databases that support delayed commits, 
a delay of one second is used.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
226 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
227 228 

<h4>Using Prepared Statements</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
229 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
230 
Wherever possible, the test cases use prepared statements.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
231 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
232 233 

<h4>Currently Not Tested: Startup Time</h4>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
234 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
235 236 237 238 239 240 241 
The startup time of a database engine is important as well for embedded use.
This time is not measured currently.
Also, not tested is the time used to create a database and open an existing database.
Here, one (wrapper) connection is opened at the start,
and for each step a new connection is opened and then closed.
That means the Open/Close time listed is for opening a connection
if the database is already in use.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
242 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
243
Thomas Mueller's avatar
--no commit message
9cf66529
 
Thomas Mueller committed
244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 
<h3>PolePosition Benchmark</h3>
<p>
The PolePosition is an open source benchmark. The algorithms are all quite simple.
It was developed / sponsored by db4o.
</p>
<table border="1" class="bar">
<tr><th>Test Case</th><th>Unit</th><th>H2</th><th>HSQLDB</th><th>MySQL</th></tr>
<tr><td>Melbourne write</td><td>ms</td><td>369</td><td>249</td><td>2022</td></tr>
<tr><td>Melbourne read</td><td>ms</td><td>47</td><td>49</td><td>93</td></tr>
<tr><td>Melbourne read_hot</td><td>ms</td><td>24</td><td>43</td><td>95</td></tr>
<tr><td>Melbourne delete</td><td>ms</td><td>147</td><td>133</td><td>176</td></tr>
<tr><td>Sepang write</td><td>ms</td><td>965</td><td>1201</td><td>3213</td></tr>
<tr><td>Sepang read</td><td>ms</td><td>765</td><td>948</td><td>3455</td></tr>
<tr><td>Sepang read_hot</td><td>ms</td><td>789</td><td>859</td><td>3563</td></tr>
<tr><td>Sepang delete</td><td>ms</td><td>1384</td><td>1596</td><td>6214</td></tr>
<tr><td>Bahrain write</td><td>ms</td><td>1186</td><td>1387</td><td>6904</td></tr>
<tr><td>Bahrain query_indexed_string</td><td>ms</td><td>336</td><td>170</td><td>693</td></tr>
<tr><td>Bahrain query_string</td><td>ms</td><td>18064</td><td>39703</td><td>41243</td></tr>
<tr><td>Bahrain query_indexed_int</td><td>ms</td><td>104</td><td>134</td><td>678</td></tr>
<tr><td>Bahrain update</td><td>ms</td><td>191</td><td>87</td><td>159</td></tr>
<tr><td>Bahrain delete</td><td>ms</td><td>1215</td><td>729</td><td>6812</td></tr>
<tr><td>Imola retrieve</td><td>ms</td><td>198</td><td>194</td><td>4036</td></tr>
<tr><td>Barcelona write</td><td>ms</td><td>413</td><td>832</td><td>3191</td></tr>
<tr><td>Barcelona read</td><td>ms</td><td>119</td><td>160</td><td>1177</td></tr>
<tr><td>Barcelona query</td><td>ms</td><td>20</td><td>5169</td><td>101</td></tr>
<tr><td>Barcelona delete</td><td>ms</td><td>388</td><td>319</td><td>3287</td></tr>
<tr><td>Total</td><td>ms</td><td>26724</td><td>53962</td><td>87112</td></tr>
</table>
Thomas Mueller's avatar
--no commit message
b492f851
 
Thomas Mueller committed
273 
<br /><a name="application_profiling"></a>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
274 275 276 
<h2>Application Profiling</h2>

<h3>Analyze First</h3>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
277 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
278 279 280 281 282 283 284 285 286 
Before trying to optimize the performance, it is important to know where the time is actually spent. 
The same is true for memory problems.
Premature or 'blind' optimization should be avoided, as it is not an efficient way to solve the problem.
There are various ways to analyze the application. In some situations it is possible to 
compare two implementations and use System.currentTimeMillis() to find out which one is faster.
But this does not work for complex applications with many modules, and for memory problems. 
A very good tool to measure both the memory and the CPU is the 
<a href="http://www.yourkit.com">YourKit Java Profiler</a>. This tool is also used
to optimize the performance and memory footprint of this database engine.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
287 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
288
Thomas Mueller's avatar
--no commit message
b492f851
 
Thomas Mueller committed
289 
<br /><a name="database_performance_tuning"></a>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
290 291 292 
<h2>Database Performance Tuning</h2>

<h3>Virus Scanners</h3>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
293 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
294 295 296 297 298 299 
Some virus scanners scan files every time they are accessed.
It is very important for performance that database files are not scanned for viruses.
The database engine does never interprets the data stored in the files as programs,
that means even if somebody would store a virus in a database file, this would
be harmless (when the virus does not run, it cannot spread).
Some virus scanners allow excluding file endings. Make sure files ending with .db are not scanned.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
300 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
301 302 

<h3>Using the Trace Options</h3>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
303 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
304 305 306 307 308 
If the main performance hot spots are in the database engine, in many cases the performance
can be optimized by creating additional indexes, or changing the schema. Sometimes the 
application does not directly generate the SQL statements, for example if an O/R mapping tool
is used. To view the SQL statements and JDBC API calls, you can use the trace options.
For more information, see <a href="features.html#trace_options">Using the Trace Options</a>.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
309 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
310 311 

<h3>Index Usage</h3>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
312 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
313 314 315 316 317 318 319 320 
This database uses indexes to improve the performance of SELECT, UPDATE and DELETE statements.
If a column is used in the WHERE clause of a query, and if an index exists on this column,
then the index can be used. Multi-column indexes are used if all or the first columns of the index are used.
Both equality lookup and range scans are supported.
Indexes are not used to order result sets: The results are sorted in memory if required.
Indexes are created automatically for primary key and unique constraints.
Indexes are also created for foreign key constraints, if required.
For other columns, indexes need to be created manually using the CREATE INDEX statement.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
321 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
322 323 

<h3>Optimizer</h3>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
324 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
325 326 327 328 329 330 
This database uses a cost based optimizer. For simple and queries and queries with medium complexity
(less than 7 tables in the join), the expected cost (running time) of all possible plans is calculated,
and the plan with the lowest cost is used. For more complex queries, the algorithm first tries
all possible combinations for the first few tables, and the remaining tables added using a greedy algorithm
(this works well for most joins). Afterwards a genetic algorithm is used to test at most 2000 distinct plans.
Only left-deep plans are evaluated.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
331 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
332 333 

<h3>Expression Optimization</h3>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
334 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
335 336 337 338 
After the statement is parsed, all expressions are simplified automatically if possible. Operations
are evaluated only once if all parameters are constant. Functions are also optimized, but only
if the function is constant (always returns the same result for the same parameter values).
If the WHERE clause is always false, then the table is not accessed at all.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
339 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
340 341 

<h3>COUNT(*) Optimization</h3>
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
342 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
343 344 345 
If the query only counts all rows of a table, then the data is not accessed.
However, this is only possible if no WHERE clause is used, that means it only works for
queries of the form SELECT COUNT(*) FROM table.
Thomas Mueller's avatar
--no commit message
81c34acc
 
Thomas Mueller committed
346 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
347 348 

<h3>Updating Optimizer Statistics / Column Selectivity</h3>
Thomas Mueller's avatar
--no commit message
2ceb8a1f
 
Thomas Mueller committed
349 
<p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
350 351 352 353 
When executing a query, at most one index per joined table can be used.
If the same table is joined multiple times, for each join only one index is used.
Example: for the query SELECT * FROM TEST T1, TEST T2 WHERE T1.NAME='A' AND T2.ID=T1.ID,
two index can be used, in this case the index on NAME for T1 and the index on ID for T2.
Thomas Mueller's avatar
--no commit message
2ceb8a1f
 
Thomas Mueller committed
354 
</p><p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
355 356 357 358 359 360 361 362 
If a table has multiple indexes, sometimes more than one index could be used.
Example: if there is a table TEST(ID, NAME, FIRSTNAME) and an index on each column,
then two indexes could be used for the query SELECT * FROM TEST WHERE NAME='A' AND FIRSTNAME='B',
the index on NAME or the index on FIRSTNAME. It is not possible to use both indexes at the same time.
Which index is used depends on the selectivity of the column. The selectivity describes the 'uniqueness' of
values in a column. A selectivity of 100 means each value appears only once, and a selectivity of 1 means
the same value appears in many or most rows. For the query above, the index on NAME should be used
if the table contains more distinct names than first names.
Thomas Mueller's avatar
--no commit message
2ceb8a1f
 
Thomas Mueller committed
363 
</p><p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
364 365 
The SQL statement ANALYZE can be used to automatically estimate the selectivity of the columns in the tables.
This command should be run from time to time to improve the query plans generated by the optimizer.
Thomas Mueller's avatar
--no commit message
2ceb8a1f
 
Thomas Mueller committed
366 
</p>
Thomas Mueller's avatar
--no commit message
23c4f01c
 
Thomas Mueller committed
367 368 

</div></td></tr></table></body></html>