linuxconsole-dev

faketty_hijackled standalone kernel patch available

[James v. Z. <ja...@dv...>]

Hi all.

I have created a standalone faketty0.05_hijackled patch downloadable
from my homepage; initially against 2.6.16.9, but it should apply
broadly.

http://members.westnet.com.au/vanzeeland

I have been getting excellent results with realtime multimedia on a 3Ghz
P4 - I am hoping to get realtime multi-console happenning, but there are
some catches. While it works fine for most applications, indeed, it's
overall latencies are down considerably, PCI video kills the mix; a
multi-console machine will be good for realtime audio only if one uses
the AGP console only, logging out all other consoles and leaving them
alone.

Any PCI console activity causes audible audio glitches. Bear in mind
that we are recording 32bit audio files from 24bit sources at 96Khz, so
there is a significantly higher load than using 16bit audio. over 4x.

I am investigating ways to drastic lower the realtime performance impact
from PCI video cards.
Is anyone aware of methods, kernel tweaks or tools that may assist in
altering the PCI performance of the PCI video?

We effectively don't care about video performance on pci consoles. If it
means they're noticably slower but audio operates in realtime, that's
great.

Does anyone have any info on how a USB console might load the system?

I will be posting realtime kernels with faketty included soon,  followed
by a more 'vanilla' kernel sometime over the next week or tso.

Apart from the audio glitches that you may not see or care about outside
of pro-audio usage, the realtime kernels provide excellent desktop
interactivity.

J
-- 
James van Zeeland <ja...@dv...>

Re: faketty_hijackled standalone kernel patch available

[Zoltan B. <zb...@fr...>]

Hi,

James van Zeeland =EDrta:
> Hi all.
>
> I have created a standalone faketty0.05_hijackled patch downloadable
> from my homepage; initially against 2.6.16.9, but it should apply
> broadly.
>
> http://members.westnet.com.au/vanzeeland
>
> I have been getting excellent results with realtime multimedia on a 3Gh=
z
> P4 - I am hoping to get realtime multi-console happenning, but there ar=
e
> some catches. While it works fine for most applications, indeed, it's
> overall latencies are down considerably, PCI video kills the mix; a
> multi-console machine will be good for realtime audio only if one uses
> the AGP console only, logging out all other consoles and leaving them
> alone.
>
> Any PCI console activity causes audible audio glitches. Bear in mind
> that we are recording 32bit audio files from 24bit sources at 96Khz, so
> there is a significantly higher load than using 16bit audio. over 4x.
>
> I am investigating ways to drastic lower the realtime performance impac=
t
> from PCI video cards.
> Is anyone aware of methods, kernel tweaks or tools that may assist in
> altering the PCI performance of the PCI video?
>  =20

Using "nice" or "renice" gives you the wanted effect.
It clearly shows on my PCI card if I start "supertux",
it skips very often but I can continue compiling on
the other console... On my system, the PCI VGA
blocks or throttles the PCI IDE in a similar way.

> We effectively don't care about video performance on pci consoles. If i=
t
> means they're noticably slower but audio operates in realtime, that's
> great.
>
> Does anyone have any info on how a USB console might load the system?
>  =20

USB used to have a CPU hog, you have to test recent kernels.

> I will be posting realtime kernels with faketty included soon,  followe=
d
> by a more 'vanilla' kernel sometime over the next week or tso.
>
> Apart from the audio glitches that you may not see or care about outsid=
e
> of pro-audio usage, the realtime kernels provide excellent desktop
> interactivity.
>
> J
>  =20

For a longer term solution, have you considered the following mainboards?
These two below are consumer-grade boards one can buy in the near future.

Gigabyte GA-G1975X, 2x PCIe x16 physical slots, 2x PCIe x4 physical
open ended slots, also usable with x16 cards (of these two, one is x1,
the other is x4 but x2 is the x1 slot is used)

Gigabyte GA-M59SLI-S5 Board, 3x PCIe x16 physical slots,
one of them is x4.

Workstation mainboard you can buy today:
Supermicro A+ *H8DCE, ***2 PCI-e x16 slots, 2 PCI-e x4 slots (x8 physical=
)

The nice thing about PCIe is that any card can work in any slot,
you just have to be able to plug them together. The card and the
slot controller negotiate the common wiring and use the highest
speed both can use. That means that one can cut the end of the slot
so an x16 card can be used in any lower grade slot, as Gigabyte
proved it with GA-G1975X.

Best regards,
Zolt=E1n b=F6sz=F6rm=E9nyi

Re: faketty_hijackled standalone kernel patch available

[James v. Z. <ja...@dv...>]

Renicing jackd to -20 and all X.org servers to 19 helps, but does not
solve the issue. I can now allow the terminals to be logged in,
expecting more than a few jack xruns, but the PCI consoles must be
abandoned when the time comes to record, as we cannot tolerate audio
glitches in the recording.

With the RRT kernel (planetccrma rt+faketty+nVidia drivers) using just
an AGP console, I get perfect performance - no xruns. at all until I
finally stopped jackd; several hours. Realtime monitoring etc turned on
the whole time, near perfect all 'round, I say.

I thought I'd gotten a good combo of IRQ's for this - are there other
suggestions out there? Anything else I can tune?

J

On Thu, 2006-05-25 at 18:19 +0200, Zoltan Boszormenyi wrote:
> Hi,
>=20
> James van Zeeland =C3=ADrta:
> > Hi all.
> >
> > I have created a standalone faketty0.05_hijackled patch downloadable
> > from my homepage; initially against 2.6.16.9, but it should apply
> > broadly.
> >
> > http://members.westnet.com.au/vanzeeland
> >
> > I have been getting excellent results with realtime multimedia on a 3=
Ghz
> > P4 - I am hoping to get realtime multi-console happenning, but there =
are
> > some catches. While it works fine for most applications, indeed, it's
> > overall latencies are down considerably, PCI video kills the mix; a
> > multi-console machine will be good for realtime audio only if one use=
s
> > the AGP console only, logging out all other consoles and leaving them
> > alone.
> >
> > Any PCI console activity causes audible audio glitches. Bear in mind
> > that we are recording 32bit audio files from 24bit sources at 96Khz, =
so
> > there is a significantly higher load than using 16bit audio. over 4x.
> >
> > I am investigating ways to drastic lower the realtime performance imp=
act
> > from PCI video cards.
> > Is anyone aware of methods, kernel tweaks or tools that may assist in
> > altering the PCI performance of the PCI video?
> >  =20
>=20
> Using "nice" or "renice" gives you the wanted effect.
> It clearly shows on my PCI card if I start "supertux",
> it skips very often but I can continue compiling on
> the other console... On my system, the PCI VGA
> blocks or throttles the PCI IDE in a similar way.
>=20
> > We effectively don't care about video performance on pci consoles. If=
 it
> > means they're noticably slower but audio operates in realtime, that's
> > great.
> >
> > Does anyone have any info on how a USB console might load the system?
> >  =20
>=20
> USB used to have a CPU hog, you have to test recent kernels.
>=20
> > I will be posting realtime kernels with faketty included soon,  follo=
wed
> > by a more 'vanilla' kernel sometime over the next week or tso.
> >
> > Apart from the audio glitches that you may not see or care about outs=
ide
> > of pro-audio usage, the realtime kernels provide excellent desktop
> > interactivity.
> >
> > J
> >  =20
>=20
> For a longer term solution, have you considered the following mainboard=
s?
> These two below are consumer-grade boards one can buy in the near futur=
e.
>=20
> Gigabyte GA-G1975X, 2x PCIe x16 physical slots, 2x PCIe x4 physical
> open ended slots, also usable with x16 cards (of these two, one is x1,
> the other is x4 but x2 is the x1 slot is used)
>=20
> Gigabyte GA-M59SLI-S5 Board, 3x PCIe x16 physical slots,
> one of them is x4.
>=20
> Workstation mainboard you can buy today:
> Supermicro A+ *H8DCE, ***2 PCI-e x16 slots, 2 PCI-e x4 slots (x8 physic=
al)
>=20
> The nice thing about PCIe is that any card can work in any slot,
> you just have to be able to plug them together. The card and the
> slot controller negotiate the common wiring and use the highest
> speed both can use. That means that one can cut the end of the slot
> so an x16 card can be used in any lower grade slot, as Gigabyte
> proved it with GA-G1975X.
>=20
> Best regards,
> Zolt=C3=A1n b=C3=B6sz=C3=B6rm=C3=A9nyi
>=20
>=20
>=20
> -------------------------------------------------------
> All the advantages of Linux Managed Hosting--Without the Cost and Risk!
> Fully trained technicians. The highest number of Red Hat certifications=
 in
> the hosting industry. Fanatical Support. Click to learn more
> http://sel.as-us.falkag.net/sel?cmd=3Dlnk&kid=107521&bid$8729&dat=12164=
2
> _______________________________________________
> Linuxconsole-dev mailing list
> Lin...@li...
> https://lists.sourceforge.net/lists/listinfo/linuxconsole-dev
--=20
James van Zeeland <ja...@dv...>

multi-console and realtime audio - up and running.

[James v. Z. <ja...@dv...>]

It seems the awnser was in the question. After retrying using an nv open
source driver for the PCI console, I have a wholly operational 2 console
realtime audio workstation.

As the open source nv driver crashes on me whenever I try it for
multi-console, I have not yet tried three or four consoles. Fedora's
x.org build has been like this for awhile. I understand others have nv
driver running multi-console; I have retained the nvidia driver on my
AGP console and this works for me.

Throwing a relatively heavy load at it, audio-wise and happy with the
result. Pity the PCI console can't handle opengl, but....I have the
multi-console workstation I was working for.

J

On Wed, 2006-05-31 at 22:17 +1000, James van Zeeland wrote:

> Renicing jackd to -20 and all X.org servers to 19 helps, but does not
> solve the issue. I can now allow the terminals to be logged in,
> expecting more than a few jack xruns, but the PCI consoles must be
> abandoned when the time comes to record, as we cannot tolerate audio
> glitches in the recording.
>=20
> With the RRT kernel (planetccrma rt+faketty+nVidia drivers) using just
> an AGP console, I get perfect performance - no xruns. at all until I
> finally stopped jackd; several hours. Realtime monitoring etc turned on
> the whole time, near perfect all 'round, I say.
>=20
> I thought I'd gotten a good combo of IRQ's for this - are there other
> suggestions out there? Anything else I can tune?
>=20
> J
>=20
> On Thu, 2006-05-25 at 18:19 +0200, Zoltan Boszormenyi wrote:
> > Hi,
> >=20
> > James van Zeeland =C3=ADrta:
> > > Hi all.
> > >
> > > I have created a standalone faketty0.05_hijackled patch downloadabl=
e
> > > from my homepage; initially against 2.6.16.9, but it should apply
> > > broadly.
> > >
> > > http://members.westnet.com.au/vanzeeland
> > >
> > > I have been getting excellent results with realtime multimedia on a=
 3Ghz
> > > P4 - I am hoping to get realtime multi-console happenning, but ther=
e are
> > > some catches. While it works fine for most applications, indeed, it=
's
> > > overall latencies are down considerably, PCI video kills the mix; a
> > > multi-console machine will be good for realtime audio only if one u=
ses
> > > the AGP console only, logging out all other consoles and leaving th=
em
> > > alone.
> > >
> > > Any PCI console activity causes audible audio glitches. Bear in min=
d
> > > that we are recording 32bit audio files from 24bit sources at 96Khz=
, so
> > > there is a significantly higher load than using 16bit audio. over 4=
x.
> > >
> > > I am investigating ways to drastic lower the realtime performance i=
mpact
> > > from PCI video cards.
> > > Is anyone aware of methods, kernel tweaks or tools that may assist =
in
> > > altering the PCI performance of the PCI video?
> > >  =20
> >=20
> > Using "nice" or "renice" gives you the wanted effect.
> > It clearly shows on my PCI card if I start "supertux",
> > it skips very often but I can continue compiling on
> > the other console... On my system, the PCI VGA
> > blocks or throttles the PCI IDE in a similar way.
> >=20
> > > We effectively don't care about video performance on pci consoles. =
If it
> > > means they're noticably slower but audio operates in realtime, that=
's
> > > great.
> > >
> > > Does anyone have any info on how a USB console might load the syste=
m?
> > >  =20
> >=20
> > USB used to have a CPU hog, you have to test recent kernels.
> >=20
> > > I will be posting realtime kernels with faketty included soon,  fol=
lowed
> > > by a more 'vanilla' kernel sometime over the next week or tso.
> > >
> > > Apart from the audio glitches that you may not see or care about ou=
tside
> > > of pro-audio usage, the realtime kernels provide excellent desktop
> > > interactivity.
> > >
> > > J
> > >  =20
> >=20
> > For a longer term solution, have you considered the following mainboa=
rds?
> > These two below are consumer-grade boards one can buy in the near fut=
ure.
> >=20
> > Gigabyte GA-G1975X, 2x PCIe x16 physical slots, 2x PCIe x4 physical
> > open ended slots, also usable with x16 cards (of these two, one is x1=
,
> > the other is x4 but x2 is the x1 slot is used)
> >=20
> > Gigabyte GA-M59SLI-S5 Board, 3x PCIe x16 physical slots,
> > one of them is x4.
> >=20
> > Workstation mainboard you can buy today:
> > Supermicro A+ *H8DCE, ***2 PCI-e x16 slots, 2 PCI-e x4 slots (x8 phys=
ical)
> >=20
> > The nice thing about PCIe is that any card can work in any slot,
> > you just have to be able to plug them together. The card and the
> > slot controller negotiate the common wiring and use the highest
> > speed both can use. That means that one can cut the end of the slot
> > so an x16 card can be used in any lower grade slot, as Gigabyte
> > proved it with GA-G1975X.
> >=20
> > Best regards,
> > Zolt=C3=A1n b=C3=B6sz=C3=B6rm=C3=A9nyi
> >=20
> >=20
> >=20
> > -------------------------------------------------------
> > All the advantages of Linux Managed Hosting--Without the Cost and Ris=
k!
> > Fully trained technicians. The highest number of Red Hat certificatio=
ns in
> > the hosting industry. Fanatical Support. Click to learn more
> > http://sel.as-us.falkag.net/sel?cmd=3Dlnk&kid=107521&bid$8729&dat=121=
642
> > _______________________________________________
> > Linuxconsole-dev mailing list
> > Lin...@li...
> > https://lists.sourceforge.net/lists/listinfo/linuxconsole-dev

--=20
James van Zeeland <ja...@dv...>