Index: chap-tuning.xml
===================================================================
RCS file: /cvsroot/htdocs/guide/en/chap-tuning.xml,v
retrieving revision 1.12
diff -b -u -r1.12 chap-tuning.xml
--- chap-tuning.xml 4 Apr 2007 08:08:05 -0000 1.12
+++ chap-tuning.xml 7 Apr 2007 10:15:17 -0000
@@ -1689,12 +1689,45 @@
<para>
LFS writes data to disk in a way that is sometimes too
- aggressive and leads to congestion.
- Information on how to throttle writing and finding the right
- parameters are available
- <ulink url="
http://mail-index.NetBSD.org/tech-perform/2007/04/01/0000.html">this</ulink> and
- <ulink url="
http://mail-index.NetBSD.org/tech-perform/2007/04/01/0001.html">this</ulink> mail.
+ aggressive and leads to congestion. Blair Sadewitz has described
+ a good method for throttling writing and finding the right
+ parameters<footnote>
+ <para>
+ See
+ <ulink url="
http://mail-index.NetBSD.org/tech-perform/2007/04/01/0000.html" /> and
+ <ulink url="
http://mail-index.NetBSD.org/tech-perform/2007/04/01/0001.html" />.
</para>
+ </footnote>.
+ The first step is to disable the sync delay, by setting the
+ following tunables through &man.sysctl.8;:
+ </para>
+
+ <programlisting>
+vfs.sync.delay = 0
+vfs.sync.filedelay = 0
+vfs.sync.metadelay = 0
+vfs.sync dirdelay = 0
+ </programlisting>
+
+ <para>
+ Then, find out how much bandwidth you have inside and outside
+ of the disk. This can be done by running
+ <command>newfs_lfs -ANF</command> on a partition close to
+ cylinder 0, and then on a partition close to the last cyclinder.
+ Average these two numbers, and calculate the LFS
+ <emphasis>page trip</emphasis> with the following formula:
+ </para>
+
+ <para>
+ <emphasis>t = (avg_bandwidth_in_bytes / PAGE_SIZE) / 4</emphasis>
+ </para>
+
+ <para>
+ Now you can set the <emphasis>vfs.lfs.pagetrip</emphasis> tunable
+ to <emphasis>t</emphasis> through &man.sysctl.8;.
+ </para>
+
+ <!-- XXX - Add note about BUFQ_PRIOCSCAN and BUFQ_READPRIO. -->
</sect2>
</sect1>