RaspyFi Rc2 – How it’s made and list of all the optimizations

Pubblicato il 14 febbraio 2013 da admin — 24 Commenti ↓

0 Flares 0 Flares ×

These optimizations come from a trial and error procedure, which led me to get an overall idea of what influences music playback performance and what doesn’t. Basically, less processes running don’t improve directly the outcome in terms of quality, but it improves overall computing performance, which indirectly affects playback quality.
Then, special attention is required to alsa optimization, using hw instead of plug or mmap_emu ensures raspyFi directly controls the Dac. So this is a big step forward.
Then, compiling the latest version of Mpd brings minor under-the-hood fixes which tend to further improve overall result.
So, this is the entire process to obtain what RaspyFi Rc2 is.

First, update to latest kernel and firmware. This brings minor usb optimizations, unfortunately the bugs affecting over 44,1 khrz playback are still present. The foundation reported to be working hard on the matter (they hired a specialist to solve the issue, at least that’s what i got…)

sudo apt-get install git-core
sudo wget http://goo.gl/1BOfJ -O /usr/bin/rpi-update && sudo chmod +x /usr/bin/rpi-update
sudo rpi-update

Then, we’ll strip down the distro. This aims basically to use less disk space, but less processes active (especially X graphic user interface) leave the pi arm cpu quieter. Some folks believe that less processes running= better sound. I am not one of those. I indeed can say that less cpu used= less power drain and more responsive system. This can indirectly improve playback quality, also for electrical inducion reasons. But my physics knowledge is not enough to explain clearly the reason why.

Sudo apt-get update

Getting a list of all packages installed give us an idea of what we can remove.

dpkg --get-selections

Let’s remove what we don’t mind to use

sudo apt-get remove aspell desktop-base desktop-file-utils dictionaries-common dillo fonts-droid fonts-freefont-ttf fontconfig fontconfig-config galculator gnome-icon-theme gnome-themes-standard gpicview gsettings-desktop-schemas gsfonts gsfonts-x11 hicolor-icon-theme leafpad lightdm lightdm-gtk-greeter lxappearance lxde lxde-common lxde-core lxde-icon-theme lxmenu-data midori mime-support openbox omxplayer penguinspuzzle x11-common wpagui python3 python3-minimal python-pygame x11-common

Python games? Glad to ask, but no.

sudo rm -rf python_games

Opt contains lot of heavy stuff, such like a damn 1080i video.

sudo rm -rf /opt/vc/src/hello_pi/hello_video/

Python, go away!

sudo apt-get remove `sudo dpkg --get-selections | grep -v "deinstall" | grep python | sed s/install//`

Removing gcc, we’ll install latest version later

sudo apt-get remove gcc-4.4-base:armhf gcc-4.5-base:armhf gcc-4.6-base:armhf

Clean unnecessary packages

sudo apt-get autoremove
sudo apt-get clean

Ok, now let’s update the software environment

sudo apt-get update
sudo apt-get upgrade

sudo apt-get clean

let’s check how much space did we free up:

df -h

Filesystem Size Used Avail Use% Mounted on
rootfs 1.8G 870M 804M 52% /
/dev/root 1.8G 870M 804M 52% /
devtmpfs 235M 0 235M 0% /dev
tmpfs 49M 252K 49M 1% /run
tmpfs 5.0M 0 5.0M 0% /run/lock
tmpfs 98M 0 98M 0% /run/shm
/dev/mmcblk0p1 56M 21M 36M 37% /boot
tmpfs 98M 0 98M 0% /tmp

870, could be worse

We don’t need swap. Do we?

sudo swapoff -a
cd /var
sudo dd if=/dev/zero of=swap bs=1M count=100

Now, we’ll install Mpd. We will now get and install the vanilla package, the one provided with debian. We will use this to get it’s infrastructure, and then we’ll install latest version from source.

sudo apt-get install mpd mpc

Now we install both prerequisites to compile mpd and mpd’s dependencies

sudo apt-get install aptitude wget binutils gcc make subversion autoconf automake autotools-dev libtool pkg-config build-essential libncurses5-dev libncursesw5-dev gcc libglib2.0-dev

sudo apt-get install libfaad2 libfaad-dev libflac8 libflac-dev libogg0 libogg-dev libvorbis0a libvorbis-dev libid3tag0 libid3tag0-dev libmad0 libmad0-dev libcue-dev libcue1 libasound2 libasound-dev libasound2-dev libao-dev libwavpack-dev libwavpack1 libsamplerate0 libsamplerate-dev libmikmod2-dev libmikmod2 libmikmod-dev libshout-dev libavformat-dev libavcodec-dev libavutil-dev libaudiofile-dev libcurl4-openssl-dev libmms-dev libmms0 libtwolame-dev libtwolame0 libmp3lame-dev git-core

Now, we clone the official mpd repo, in order to obtain and compile latest mpd version. As root:

sudo su
cd /home/pi
git clone git://git.musicpd.org/master/mpd.git

cd ./mpd

./autogen.sh

We then configure it to use few options, making mpd more lightweight.

./configure --disable-bzip2 --disable-iso9660 --disable-zzip --enable-id3 --disable-sqlite --enable-ffmpeg --enable-alsa --disable-wave-encoder --enable-pipe-output --enable-httpd-output --disable-recorder-output --disable-sndfile --enable-oss --enable-shout --disable-pulse --disable-ao --disable-mad --disable-inotify --disable-ipv6 --enable-curl --disable-mms --disable-wavpack --disable-lame-encoder --disable-twolame-encoder --enable-vorbis --enable-lsr --with-zeroconf=auto

make
make install
sudo su
/etc/init.d/mpd stop
echo DAEMON=/usr/local/bin/mpd >>/etc/default/mpd
/etc/init.d/mpd start
mkdir -p ~/.mpd/playlists

Now, we’ll start configuring mpd, editing the mpd.conf file

sudo nano /etc/mpd.conf

What interest us is:

#group "nogroup"

Becomes

group "audio"

Mpd will belong to “audio” group, which will be useful for setting mpd’s priorities.

Then

bind_to_address "localhost"

that will become

bind_to_address "any"

Activating auto-update by uncommenting following line, this will enable mpd to automatically update its music database when something new is added or removed.

#auto_update "yes"

Then

audio_output {
type "alsa"
name "My ALSA Device"
device "hw:0,0" # optional
format "44100:16:2" # optional
mixer_device "default" # optional
mixer_control "PCM" # optional
mixer_index "0" # optional

that will become

audio_output {
type "alsa"
name "USB Dac"
device "hw:0,0" # optional
format "44100:16:2" # optional
# mixer_device "default" # optional
# mixer_control "PCM" # optional
# mixer_index "0" # optional

Then we’ll uncomment the lines regarding the buffer

audio_buffer_size "2048"
buffer_before_play "10%"

Ctrl+ x to save, then y to confirm.

Now we assign to our usb dac the device number 0, which is the default for mpd and the mixer. We can do so by commenting out the line

options snd-usb-audio index=-2

sudo nano /etc/modprobe.d/alsa-base.conf

By doing so, RaspyFi will use as default the USB Dac. If a usb dac is not connected, then the default audio out will be the analog jack. Nasty!

Now, we create our music directories and mount points. We can use multiple directories, such a nas and a usb drive, by putting a shortcut in mpd default music directory.

mkdir /mnt/nas
chmod 777 /mnt/nas

Add the following line in /etc/fstab

nano /etc/fstab

//192.168.1.NasIP/YourShareName /mnt/nas cifs username=pi,password=,uid=mpd,file_mode=0644,dir_mode=0755,iocharset=utf8,rsize=130048,wsize=4096 0 0

This will automatically mount our samba share on /mnt/nas.

Now we mount an usb drivE. Please note that it is preferrable to use a fat formatted drive, this uses less resources. This is quite important, considering the notorious usb problems.

See how the drive is called

sudo fdisk -l

in my case:
/dev/sda1

We create the music folder:

mkdir /media/music
chmod -R 777 /media/music
mount /dev/sda1 /media/music

Then we add this line to /etc/fstab

sudo nano /etc/fstab

/dev/sda1 /media/music vfat defaults 0 0

Ctrl+ X then, y
We now add a symlink into the default music directory, this will give us the possibility to store our music library on multiple locations. For example, we can have a part of our library on the nas and another one on a usb drive

cd /var/lib/mpd/music
sudo ln -s /mnt/nas
sudo ln -s /media/music

Installing wi-fi client:

sudo apt-get install wicd-curses

cd /var/lib/mpd/music/

Tweaking

This is a fine tuning of the Pi software echosystem, even if we cannot get rid (as for now) of the usb problems, we can still do something to improve general usb performance. Adding this line to:

sudo nano /etc/modprobe.d/alsa-base.conf

options snd-usb-audio nrpacks=1

This sounds really effective to my hears!

There’s another tweak that should fix the usb problems, you should add a line in the cmdline.txt. This line turns the usb into 1.1 mode. Please note that by doing so, both usb and ethernet performance in terms of transfer speed will be greatly reduced. This doesn’t work for me, as my Usb Dac doesn’t work with this mode. Feel free to try it, please report if it solves the problem with your dac. You can remove this tweak by deleting the added parameter.

sudo nano /boot/cmdline.txt

dwc_otg.speed=1

Next step is to set the Pi’s cpu to work always at its full power. This is due to the fact that mpd for an unknown reason doesn’t trigger the scaling governor to increase computing speed while in use. By enabling the “performance” option, we’ll be sure that mpd has the full power it deserves.
First, we check out which power management profile we have:

cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
powersave

Ok, let’s put the Pi in performance mode:

sudo su
echo -n performance > /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor

Check again with previous command to see if it is effective.
Using hw, insted of plug and mmemu

nano /etc/asound.conf

cancel existing text and paste this

pcm.!default {
type hw
card 0
}

ctl.!default {
type hw
card 0
}

LAN TUNING

sudo nano /etc/sysctl.conf

add line

fs.inotify.max_user_watches = 524288

If you should have some problems, remove it. In my configuration I don’t have any problems, but sometimes they could occour depending on traffic and tcp tranfer size.

Adjust limit values of the audio group

First we add mpd to the audio group

sudo nano /etc/security/limits.conf

@audio - rtprio 99
@audio - memlock unlimited
@audio - nice -19

Then we set mpd to have the highest priority (niceness). We do so by adding a script that starts afer mpd that renices it to -20

sudo su
cd /etc/init.d
nano myruns

We put this in the script:

#! /bin/sh
# /etc/init.d/myruns
#
### BEGIN INIT INFO
# Provides: myruns
# Required-Start: $all
# Required-Stop: $mpd
# Default-Start: 2 3 4 5
# Default-Stop: 0 1 6
# Short-Description: Renicing mpd
# Description: Being nice with music
### END INIT INFO
# Some things that run always
touch /var/lock/myruns

MAXTRIES=15
TRIES=0

# Carry out specific functions when asked to by the system
case "$1" in
start)
#
# mount music
#
renice -n -20 -p `pgrep mpd`
;;
stop)
echo "Exiting myruns"
;;
*)
echo "Usage: /etc/init.d/myruns {start|stop}"
exit 1
;;
esac

exit 0
Ctrl+ x then y

We then set this as script and we add it to the init.d boot sequence:

chmod 755 myruns
update-rc.d myruns defaults

NAS tuning

sudo nano /etc/fstab

adding line
uid=mpd,file_mode=0644,dir_mode=0755,iocharset=utf8,rsize=130048,wsize=4096

Now we selectively deactivate some process, so they don’t start up nad we’ll have less active processes.

sudo apt-get install rccconf
sudo rccconf

The services I disabled:

cron – scheduler
lightdm – graphic login manager
motd – message of the day for ssh
plymouth- graphic bootlogo
x11-common graphic interface

This broke somehow my ssh server, avoiding me to connect via ssh, so be aware of it. If this will happen you should have access to your pi via a monitor and a keyboard and reinstall sshd via those commands:

sudo apt-get remove openssh-client openssh-server
sudo apt-get install openssh-client openssh-server

Now, we proceed to remove everything we used to compile and several things more. In case we don’t think we’re going to recompile something again.

sudo apt-get remove autoconf automake autotools-dev binutils build-essential dpkg-dev g++ g++-4.6 gcc gcc-4.6 libglib2.0-dev libid3tag0-dev libsamplerate0-dev libstdc++6-4.6-dev make pkg-config git-core x11-common dbus-x11 libice6 libx11-6 libx11-data libx11-xcb1 libxext6 libxi6 libxmuu1 libxtst6 xauth xkb-data fonts-freefont-ttf

cd /home/pi
sudo rm -rf mpd

That’s all folks, shake and serve with ice!

Taggato con: audio quality, install, Mpd, optimizations, raspyfi, tutorial, usb dac
Pubblicato in Documentation, Tutorials

cug

I think this is the sixth time I follow these instructions (I keep messing with raspbian until I need to install it again and its main purpose is music with cmus) Thanks a lot for your contribution
Fred

I’ve just installed raspyfi 1.0 . Nice work, I get the feeling you’ve created something many of us desired.

Though I have this question:

“Volume control Mixer / Hardware” kinda works, it adjusts the first channel for my USB DAC, which is BASS level :

[email protected]:~$ amixer controls
numid=4,iface=MIXER,name=’Tone Control – Bass’
numid=5,iface=MIXER,name=’Tone Control – Treble’
numid=2,iface=MIXER,name=’PCM Playback Switch’
numid=3,iface=MIXER,name=’PCM Playback Volume’
numid=6,iface=MIXER,name=’Line Capture Volume’
numid=1,iface=PCM,name=’Playback Channel Map’

Any idea how I can tell mpd to adjust numid == 3 instead of 4 ?

Thanks!
- Fred
  
  Ahhh, found my answer :
  
  /etc/mpd.conf:
  audio_output {…}
  mixer_control = PCM
  
  Thanks again.
D

Love this product! Any news as to when an updated release coming?
Toby

Is there any chance of getting this RC2 in the updated version for download ?

And when is the new version coming out? I can’t figure out how to fix my Raspify problems.

After setting everything up, it works for the time being. As soon as I turn it off, the web UI doesn’t load anymore and the PI is listed in the network, but doesn’t respond. =(
Vincent Li

Hello Jeremy, thanks for all the tweak tips. I’ve had a play around earlier with the Raspberry pi but gave up with the USB issues. In my adventures one of the biggest improvements I got was with the real-time kernel – it reduced the noise floor significantly compared to the standard (pre-empt?) kernel (although I don’t remember the kernel version this was over a year ago). If you haven’t yet tried this I recommend you to try it. Its called the rt kernel (not the pre-empt one!). Hope this helps!
- AutoStatic
  
  Hi Vincent, you should thank the author of this article (mikelangeloz) and not me And using a RT kernel does not affect audio in any way, that’s an audiophile myth.
  
  Jeremy
  - Vincent Li
    
    lol Jeremy thanks for pointing out the real author I only had a quick look but didn’t see it in the article itself. I’m a bit surprised to hear that you find the RT kernel does not make any improvement to the audio – for me I did a controlled experiment keeping all the variables the same and found that it significantly reduced the frequency of the pops from every few seconds to every minute or so and the background noise was much lower.
Pingback: Yet Another Musical (Raspberry) Pi « Mark's Pages Of Stuff
Andrew McGregor

thanks for a good write-up.
A couple of suggestions. I would add “–prefix=/ –exec-prefix=/usr/” to the “configure command” above to get a compiled mpd version installed on top vanilla installed mpd. (don’t see a point to keep two copies of mpd in the system). therefore – no need to modify /etc/default/mpd .
Another thing to simply the install – it’s easier to add “–nicelevel” agrument to /etc/init.d/mpd instead of creating a new script to change nice level for mpd. I.e:

start-stop-daemon –nicelevel $NICE_LEVEL –start –quiet –oknodo –pidfile “$PIDFILE”

–exec “$DAEMON” — $MPD_OPTS “$MPDCONF”

$NICE_LEVEL var has to be defined somewhere above like so:

NICE_LEVEL=-19

One more thing if you need flac support you would need to install libflac-dev and add –enable-flac to the “configure command”
- mikelangeloz
  
  Thanks for the clever comment! You’re right, no point of having 2 versions of MPD installed, in the new version of RaspyFi there will be only one.
  As for the renice in the init.d script, thanks! This is a proper solution, gonna implement this in 1.0!
  As for flac, it’s implemented!
uvxx

hiface two worked just once!

It worked at the time I built the system and installed audacious,and never work again,no matter how I set the preferences in audacious

and the sound is bad，Sonic boom
- admin
  
  Audacious? Why? To control mpd?
John Fro

When I do this, mpd just won’t start, either on boot or when init.d is called manually. Not sure where to look.
- michelangelo
  
  Copy the message you get, and we try to solve it
Christ

Hi there,

First, thank you very much for your effort. I am running the image atm. and find it to have a lot better performance for hd tracks than the std. image.

I have a few suggestions.

First, you might want to clear the MPD DB before making the image, so people dont get your collection upon first load.

Second, comment out this part in MPD.conf, to achieve bitperfect communication to an external DAC,

format “44100:16:2″ # optional

If not, tracks at another bitrate will be down or upsampled.

Again, thanks a lot for your work, i will follow it with great interest.

Cheers

Christian
- admin
  
  Hello Christian, thanks for your feedbacks and suggestions. You’re right about clearing the db.
  Regarding the resampling, this is necessary. The Raspberry Pi cannot handle (by now) high data transfers.
  Files above 48 khrtz and 16 bit sounds very distorted, this is due to packet losses. These problems are to be solved by the foundation, because the usb driver is very buggy.
  - Christ
    
    With my setup, i play 24/96 with no pops or garbled sounds if i play from a direct attached USB disk.
    
    From NAS it is not possible.
    - admin
      
      Could you please tell me which dac are you using?
      Have you enabled the dwg_otg line?
      - Christ
        
        Hi again,
        
        My signal chain when uising the PI is,
        
        Pi -> Hiface2 -> Hegel HD20
        
        With the RaspyFI image, i played a few hours of 24/96 last night, with not a single pop.
        
        I did however try and use the USB input of my Hegel, and found that the sound was once again garbled.
        
        For some reason, the Hiface2, seems to work…. i must admit i am clueless to why this is.
        
        I do not have the dwg_otg line set, since the Hiface2 is a USB2 device, and stops working when i try it.
        
        Hope this helps
        
        Christian
admin

Hello, Jeremy.
You are right about the nrpacks option, there’s however still a debate on that being useful or not. So I decided to put it anyway, better safe than sorry…
The line you mentioned is meant to improve lan transfers performance by limiting the lenght of paths stored by the kernel during i\o activities. However, I’m not able to say if it brings some audible improvement.
Do you have any suggestion for further improvements?
- http://wiki.linuxaudio.org AutoStatic
  
  You’re the first one I hear saying something sensible about the inotify stuff It’s related to performance and won’t bring any audible improvement. I highly doubt if it makes any difference though performance wise. But it could be different on the RPi as it’s much less powerful than a desktop or notebook.
  
  As for suggestions, you might want to check:
  * http://wiki.linuxaudio.org/wiki/raspberrypi
  * http://wiki.linuxmusicians.com/doku.php?id=system_configuration
  * If you have time and can bear with someone uttering a lot of eeeermmmms: http://lac.linuxaudio.org/2011/video.php?id=3
  I don’t have any direct suggestions regarding Raspyfi as I haven’t tried it (yet).
  - admin
    
    Hi Jeremy, nice to hear we share same convinctions!
    Thanks for your links, i found some useful things to further improve the RaspyFi environment, as for the conference, now I know what to do tonight before sleep.
    In my opinion, the inotify stuff has some advantages for the Raspberry Pi, as minor stress is given on the usb bus (which shares both usb and ethernet).
    As you may know, the usb bus is the real bottleneck of the Raspberry Pi. I’m waiting for the Foundation to solve the issue, meanwhile I’m about to connect the Raspberry to a Dac via i2s through the i2s pins.
    Nice to hear you in these pages, hope you’ll be back soon!
    - http://linux.autostatic.com/ AutoStatic
      
      yes, USB is a bit of a pita on the RPi. I’ve disabled the turbo mode of the ethernet controller and that already improved stability a lot. Another option would be to disable the ethernet controller altogether but then you’d have no more SSH access anymore.
      Is there already a DAC via i2s available?
http://linux.autostatic.com/ AutoStatic

Isn’t using the nrpacks option deprecated? See: http://lists.linuxaudio.org/pipermail/linux-audio-user/2010-January/066765.html
So I think it actually does nothing unless the USB host controller on the RPi uses a specific driver that does have this option.
Regarding using nice, most RPi distros have a PREEMPT kernel which allows you to effectively use chrt. In an audio context this is the preferred way to prioritize processes.
And I’d really like to know what the following line does and how it affects performance:
fs.inotify.max_user_watches = 524288

Regards,

Jeremy