Mercurial > hg > nginx-site

comparison xml/en/docs/freebsd_tuning.xml @ 747:ddec437b692b

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Fixed a number of typos.
author Vladimir Homutov <vl@nginx.com>
date Wed, 31 Oct 2012 09:18:29 +0000
parents 764fbac1b8b4
children 9ca8b71e0e03
comparison
equal deleted inserted replaced
746:a02360a5c265 747:ddec437b692b
117 You may change them on the fly using the sysctls 117 You may change them on the fly using the sysctls
118 <path>net.inet.tcp.recvspace</path> and 118 <path>net.inet.tcp.recvspace</path> and
119 <path>net.inet.tcp.sendspace</path>. 119 <path>net.inet.tcp.sendspace</path>.
120 Of course the bigger buffer sizes may increase throughput, 120 Of course the bigger buffer sizes may increase throughput,
121 because connections may use bigger TCP sliding windows sizes. 121 because connections may use bigger TCP sliding windows sizes.
122 And on the Internet you may see recomendations to increase 122 And on the Internet you may see recommendations to increase
123 the buffer sizes to one or even several megabytes. 123 the buffer sizes to one or even several megabytes.
124 However, such large buffer sizes are suitable for local networks 124 However, such large buffer sizes are suitable for local networks
125 or for networks under your control. 125 or for networks under your control.
126 Since on the Internet a slow modem client may ask a large file 126 Since on the Internet a slow modem client may ask a large file
127 and then it will download the file during several minutes if not hours. 127 and then it will download the file during several minutes if not hours.
135 The server may place a whole large response in the sending buffer, then may 135 The server may place a whole large response in the sending buffer, then may
136 close the connection, and let the kernel to send the response to a slow client, 136 close the connection, and let the kernel to send the response to a slow client,
137 while the server is ready to serve other requests. 137 while the server is ready to serve other requests.
138 You should decide what is it better to bind to a client in your case: 138 You should decide what is it better to bind to a client in your case:
139 a tens megabytes Apache/mod_perl process 139 a tens megabytes Apache/mod_perl process
140 or the hundreds kilbytes socket sending buffer. 140 or the hundreds kilobytes socket sending buffer.
141 Note that nginx uses non-blocking I/O system calls 141 Note that nginx uses non-blocking I/O system calls
142 and devotes just tens kilobytes to connections, 142 and devotes just tens kilobytes to connections,
143 therefore it does not require the large buffer sizes. 143 therefore it does not require the large buffer sizes.
144 </para> 144 </para>
145 145
217 217
218 <note> 218 <note>
219 The page size jumbo clusters have been introduced in FreeBSD 7.0. 219 The page size jumbo clusters have been introduced in FreeBSD 7.0.
220 In earlier versions you should tune only 2K mbuf clusters. 220 In earlier versions you should tune only 2K mbuf clusters.
221 Prior to FreeBSD 6.2, the <path>kern.ipc.nmbclusters</path> value can be 221 Prior to FreeBSD 6.2, the <path>kern.ipc.nmbclusters</path> value can be
222 set only on the boot time via loader tunnable. 222 set only on the boot time via loader tunable.
223 </note> 223 </note>
224 </para> 224 </para>
225 225
226 <para> 226 <para>
227 On the amd64 architecture FreeBSD kernel can use for sockets buffers 227 On the amd64 architecture FreeBSD kernel can use for sockets buffers
228 almost all physical memory, 228 almost all physical memory,
229 while on the i386 architecture no more than 2G memory can be used, 229 while on the i386 architecture no more than 2G memory can be used,
230 regardless of the available physical memory. 230 regardless of the available physical memory.
231 We will discuss the i386 specific tunning later. 231 We will discuss the i386 specific tuning later.
232 </para> 232 </para>
233 233
234 <para> 234 <para>
235 There is way not to use the jumbo clusters while serving static files: 235 There is way not to use the jumbo clusters while serving static files:
236 the <i>sendfile()</i> system call. 236 the <i>sendfile()</i> system call.
354 </para> 354 </para>
355 355
356 </section> 356 </section>
357 357
358 358
359 <section id="minor_optmizations" 359 <section id="minor_optimizations"
360 name="Minor optimizations"> 360 name="Minor optimizations">
361 361
362 <para> 362 <para>
363 363
364 <programlisting> 364 <programlisting>