[LAU] jack2 vs. jack1/zita vs. jack2+zita-a2j, performance differences

Previous message: [thread] [date] [author]
Next message: [thread] [date] [author]
To: linux-audio-user <linux-audio-user@...>
Date: Thursday, April 17, 2014 - 1:54 am

------=_Part_75113_1885810296.1397699585072
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit

Thought the below might be of interest to some. The last is the best :-) Hardware and clients identical, octo 4GHz, test load is one yoshimi with jack_keyboard driving by way of a passthrough mididings.

----------------------------

Item one is vanilla distro jack2, run like this:

jack_control start
jack_control ds alsa
jack_control dps device hw:SB
jack_control dps rate 48000
jack_control dps nperiods 2
jack_control dps period 128
jack_control dps midi-driver seq
jack_control dps inchannels 2
jack_control dps outchannels 2
jack_control eps realtime true
jack_control eps realtime-priority 50
jack_control eps clock-source 1

The above is as low in reported latency I could go, 5.33ms, using every optimization I know of via the Arch wiki and realtimeconfigquickscan and other sources. Running load-less, it gives me usage ranging 0.9 through 1.4%, usually hanging in at 1.3%. With the test load, it tends to sit at 10% and rise with stress. A few xruns every once in a while are inevitable, even when the test load is delivering only silence. Nice tone quality.

-----------------------------

Item two is jack1-git, compiled with Zita libraries engaged, run like this:

nohup schedtool -R -p 50 -e /usr/bin/jackd -A SB -R -c h -X alsa_midi -d dummy -r 48000 -p 32 &

Reported latency is 2ms. Load-less, usage rating is 1.3% through 29%, usually hanging in at 1.4% or so. With the test load, it sits at 31.2% at silence. No xruns. But no actual sound came out :-)

-----------------------------

Item three is jack1-git, Zita libraries present but not used, run like this:
nohup schedtool -R -p 50 -e /usr/bin/jackd -R -c h -X alsa_midi -d dummy -r 48000 -p 32 &
nohup schedtool -R -p 50 -e /usr/local/bin/zita-j2a -d hw:SB

2ms reported latency. Loadless usage is 20%. With test load, 25-33% at silence. No xruns. Distorted sound.

------------------------------

Item four is jack2 vanilla distro, run like this:

jack_control start
jack_control ds dummy
jack_control dps rate 48000
jack_control dps nperiods 2
jack_control dps period 32
jack_control dps inchannels 2
jack_control dps outchannels 2
jack_control eps realtime true
jack_control eps realtime-priority 50
jack_control eps clock-source 1

schedtool -R -p 50 -e /usr/bin/a2jmidid
schedtool -R -p 50 -e /usr/local/bin/zita-j2a -d hw:SB

2ms reported latency. Loadless usage 3.8 to 4.3%. With triple test load (three Yoshimis which is actually SOP for this rig), 4.5-5.5% at silence. No xruns. Wonderful sound, and no xruns with that triple load being pushed.

Will be starting rigorous testing at lower latencies soon, but I have an instrument which has to sing first :-)

Jonathan E. Brickman
Ponderworthy Music | jeb@ponderworthy.com | (785)233-9977 | http://ponderworthy.com

------=_Part_75113_1885810296.1397699585072
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

=
Thought the below might be of interest to some.  The last is the=
best :-)  Hardware and clients identical, octo 4GHz, test load is one=
yoshimi with jack_keyboard driving by way of a passthrough mididings.<=
div>----------------------------It=
em one is vanilla distro jack2, run like this:jack_control startjack_contro=
l ds alsajack_control dps device hw:SBjack_con=
trol dps rate 48000jack_control dps nperiods 2jack_co=
ntrol dps period 128jack_control dps midi-driver seqj=
ack_control dps inchannels 2jack_control dps outchannels 2=
jack_control eps realtime truejack_control eps realtime-pri=
ority 50jack_control eps clock-source 1The above is as low in reported latency I could go, 5.33ms, using ev=
ery optimization I know of via the Arch wiki and realtimeconfigquickscan an=
d other sources.  Running load-less, it gives me usage ranging 0.9 thr=
ough 1.4%, usually hanging in at 1.3%.  With the test load, it tends t=
o sit at 10% and rise with stress.  A few xruns every once in a while =
are inevitable, even when the test load is delivering only silence.  N=
ice tone quality.-----------------------------Item two is jack1-git, compiled with Zita libraries =
engaged, run like this:nohup schedtool -R -p 50 -=
e /usr/bin/jackd -A SB -R -c h -X alsa_midi -d dummy -r 48000 -p 32 &Reported latency is 2ms.  Load-less, usage rat=
ing is 1.3% through 29%, usually hanging in at 1.4% or so.  With the t=
est load, it sits at 31.2% at silence.  No xruns.  But no actual =
sound came out :-)-----------------------------Item three is jack1-git, Zita libraries present but=
not used, run like this:nohup schedtool -R -p 50 -e /usr/bin/ja=
ckd -R -c h -X alsa_midi -d dummy -r 48000 -p 32 &nohup sche=
dtool -R -p 50 -e /usr/local/bin/zita-j2a -d hw:SB2ms reported latency.  Loadless usage is 20%.  With test load, 2=
5-33% at silence.  No xruns.  Distorted sound.------------------------------Item four is=
jack2 vanilla distro, run like this:jack_co=
ntrol startjack_control ds dumm=
yjack_control dps rate 4=
8000jack_control dps nperiods 2jack_control dp=
s period 32jack_control dps inc=
hannels 2jack_control dps outchannels 2jack_co=
ntrol eps realtime truejack_control eps realtime-priority 50jack_control eps clock-source 1schedtool -R -p 50 -=
e /usr/bin/a2jmididschedtool -R -p 50 -e /usr/local/bin/zita-j2a=
-d hw:SB2ms reported latency.  Loadless usa=
ge 3.8 to 4.3%.  With triple test load (three Yoshimis which is actual=
ly SOP for this rig), 4.5-5.5% at silence.  No xruns.  Wonderful =
sound, and no xruns with that triple load being pushed.Will be starting rigorous testing at lower latencies soon, but I have=
an instrument which has to sing first :-)Jonathan E. BrickmanPonderworthy Music =
| jeb@ponderworthy.com | (785)233-9977 | http://ponderworthy.com
------=_Part_75113_1885810296.1397699585072--

Previous message: [thread] [date] [author]
Next message: [thread] [date] [author]

Messages in current thread:
[LAU] jack2 vs. jack1/zita vs. jack2+zita-a2j, performance d..., Jonathan E. Brickman, (Thu Apr 17, 1:54 am)
Re: [LAU] jack2 vs. jack1/zita vs. jack2+zita-a2j, performan..., Jonathan E. Brickman, (Thu Apr 17, 12:36 pm)