Differences

This shows you the differences between two versions of the page.

--- internet:self-hosting [2011/12/22 09:39]
cyril [Flash cache]
+++ internet:self-hosting [2013/09/19 16:40] (current)
@@ Line 1: / Line 1: @@
 ====== Website Self Hosting ======
+===== IP =====
+Dynamic DNS service if you don't have a static IP.
 ===== Hardware =====
@@ Line 6: / Line 10: @@
 ==== Flash cache ====
+=== Introduction ===
+Solutions for hybriding flash drives and hard drives have been experimented to save power on laptops, but it is quite complex in the general case and there is no final-user implementation yet, and the gains are limited on an everyday use machine. However it is a lot more efficient and simple on a server, as there are a lot less interactions with users and the panel of activities is largely reduced. Hence you only have to put on flash disk a few directories and files that the system regularly write to (the cache in RAM is enough for reading), and the disk can remain spun down for hours or days if you don't start unusual applications.
-A lot more efficient and simple on a server than on a work machine.
+You need an USB stick or SD Card of at least 2GB (4GB is advised), and an init script to mount and fetch during early boot process, and unmount and flush during late shutdown process. For instance I wrote this one for Gentoo: [[init_fcache|/etc/init.d/fcache]] with conf [[conf_fcache|/etc/conf.d/fcache]], add it to the boot runlevel (''rc-update add fcache boot'').
-<code>
+As an example, my server (a Core2 Duo laptop) consumes 12-13W (saved 1W stopping the hard drive), makes no noise at all (no fan, no hard drive), and the hard drive remains stopped for several days, as long as I don't interact with because it is also a multimedia server (I can even log in with ssh, with export display, as root, use firefox, without waking up the drive ; but of course listening to music stored on the hard drive, installing a program, changing some system configuration, starting an unusual program, will wake up the drive).
-/var/log
-/var/www
-/var/lib
-/var/cache ?
-/var/db ?
-ou tout /var... un peu gros mais ça peut passer
-/tmp (ou en ram)
-</code>
+=== What to cache ===
-Create ~/.fcache dir, move all files you want to cache to it, and either create a symbolic link or configure the program to use this new location:
+The main directories you need to take care of are:
-<code>
+  * ''/var'' must be cached
-~/.history
+  * ''/tmp'' must be mounted as tmpfs in ram
-~/.xsession-errors
+  * ''/home'' can be cached if you have enough space on the flash disk
+If you cannot cache your whole home folder, you can cache a few files that are updated too often by putting them in a ''~/.fcache'' directory that you will cache. Then either create a symbolic link (if it works, meaning that the file is not deleted and recreated each time) or configure the program to use this new location:
+  * ''~/.history'': shell history for all users including root; symlink not ok, set the new location with the $HISTFILE environment variable for zsh in .zshrc file.
+  * ''~/.lesshst'': ''less'' history; symlink ok.
+  * ''~/.xsession-errors'': gdm file; symlink not ok, needs version >2.32.1 so that symlink works, with previous versions you have to recompile it after hard changing ".xsession-errors" to ".fcache/.xsession-errors" (in file ''daemon/slave.c'' for version 2.20).
+  * ''~/.Xauthority'': xauth file; symlink not ok:
+    * for GDM: set ''[daemon] UserAuthDir=/home/cyril/.fcache'' in ''/etc/X11/gdm/custom.conf''
+    * for SSHD: set ''XAUTHORITY=/home/cyril/.fcache/.Xauthority'' in ''~/.ssh/environment'', and ''PermitUserEnvironment yes'' in ''/etc/ssh/sshd_config''
+    * for SU: set ''cyril'' (your username) in ''/root/.xauth/import'', to avoid creation of ''/root/.xauthXXXXX'' files
+  * application data folder of programs you commonly use (firefox, music player, ...)
+''/etc'' cannot be cached entirely, because it contains the init scripts that are used to setup fcache (''/etc/init.d''), but there are some files that are necessary to cache:
+  * ''/etc/resolv.conf'': symlink ok.
+=== Configure monitors ===
+If you run ''smartd'' daemon, it will wakes up the drive every 30min by default to perform some checks (or with the interval specified by the ''-i'' option of ''smartd''). You have to tell it to not do tests if the drive is in standby powermode. Edit ''/etc/smartd.conf'':
+<code diff>
+-DEVICESCAN
++#DEVICESCAN
++/dev/sda -d ata -n standby
 </code>
+Asking the temperature to the ''hddtemp'' daemon will also wake up the drive, so you should not use it.
+Also be careful to conflicts between manual configuration and different power manager utilities (like pm-utils, your desktop environment...). You may have to disable some of them if it doesn't work.
+=== Finding the guilty files ===
+You can use a script similar to this one to monitor what files are being used and add them to the cache (you can then grep "sda" and "dirtied"), and monitor if the drive remains spun down:
 <code sh>
 #!/bin/sh
+rm -f /mnt/ram/disk.log
 dmesg -c
-rm -f /mnt/ram/disklog
+#sudo sysctl vm.block_dump=1
 echo 1 > /proc/sys/vm/block_dump
-watch "dmesg -c >> /mnt/ram/disklog"
+watch --interval=5 'dmesg -c | grep sda >> /mnt/ram/disk.log ; hdparm -C /dev/sda | grep "drive state" | cut -d":" -f2 >> /mnt/ram/disk.log'
 echo 0 > /proc/sys/vm/block_dump
 </code>
+You can also use ''strace -p <PID> -e trace=write -s 128'' if you don't have enough information about some files name and location.
-<code sh>
+=== Configure the drive to go to standby ===
-hdparm -C /dev/sda
+Configure hdparm to put the drive to standby after some time of inactivity. For instance to test with 10 minutes:
+<code>
+hdparm -S 120 /dev/sda
 </code>
+And for final configuration copy ''/usr/lib/pm-utils/power.d/harddrive'' file to ''/etc/pm/power.d/'' and set variable ''DRIVE_SPINDOWN_VALUE_BAT'' to 241 for instance for 30 minutes (check option -S in hdparm's man page).
+You can also configure with your power manager utility if you have one.
+=== Optional extra configuration ===
+You can try to tune how syslog will write to the disk to save the flash life, in ''/etc/syslog-ng/syslog-ng.conf'':
+<code>
+options {
+    mark_freq(0); # remove --MARK-- lines issued by default every 20s
+    flush_lines(20);
+    flush_timeout(60000); # in ms
+};
+</code>
+=== Read cache ===
+You can try to prefetch in the Linux RAM read cache all the programs and files you often read, with a script that executes programs and grep inside files when you start the computer.
+You can also have a look at ''e4rat'' that optimizes the boot process but also prefetches files used at boot, and ''preload'' that adaptatively preloads the programs you usually use.
+=== RAM write cache ===
+You could also try to use the ''noflushd'' daemon and ''laptop-mode-tool'' to tune the RAM write cache (how long and how much before flushing to the disk), but if you lose power you may lose lots of data (that were recently written), and it may not work as expected because of the filesystem that force some writes. See [[http://buffalo.nas-central.org/wiki/Spin_down_the_hard_drive_when_idle]].