Mailing List Archive

Draft revisions to HowTo section 3.3, second try
Based on this morning's feedback -- both directly to my first draft and on
other issues that were discussed in the last day -- I've revised the draft
update a good bit. Here is the new version, for comments and criticisms.

Thank you to everyone who offered suggestions. I invite additional
feedback, although my incorporating them into a third draft will probably
wait until I get some feedback from Robert on the baseline merits of doing
this revision.


--------------------DRAFT BEGINS---------------------

3.3 Hardware

Hardware selection is a complex topic, one this HowTo can touch on only
briefly and in general terms. The following subsections offer some general
guidence but stop short of offering specific recommendations.

For more detail about actual configurations that others have used, Mark
Cooper has setup a hardware database at
http//www.goldfish.org/~mcooper/pvrhw/. The website will let you browse
what other users have reported as their hardware configuration, and how
happy they are with the results.

To find out more about the suitability of specific hardware choices, you
can also consult the archives of the mythtv-users mailing list, or post a
question on that list.


3.3.1 CPU Type and Speed

Selection of CPU type and speed is the trickiest element of hardware
selection, mainly because there are so many tradeoffs one can make, among
number of things the MythTV device can do simultaneously, capture size, and
encoding quality.

MythTV has two modes of operation. It can function as a software video
encoder, which means that it uses a fairly generic "dumb" video capture
card to get frames of video, encodes them using the CPU on your motherboard
and writes them to disk. High-end video capture cards and devices like the
Tivo and RePlay have dedicated encoder chips which use specialized hardware
to convert the video stream to the MPEG-2 format without using the
motherboard CPU. The main CPU has the responsibility of running the
Operating System and reading and writing the encoded frames to the disk.
These tasks have fairly low CPU requirements compared to encoding video,
which is why a device like a Series 1 Tivo can run with only 16MB of RAM
and a 54Mhz CPU.

There are many variables that go into the question "How fast a CPU do I
need to run MythTV"? Obviously, the faster your CPU, the better your
experience will be with MythTV. If you are using the software MPEG-4
encoder and performing the "Watch TV" function, where the CPU is both
encoding and decoding video simultaneously to allow Pause, Fast Forward and
Rewind functions for live TV requires more CPU then just encoding or
decoding. MythTV also supports multiple encoder cards in a single PC,
thereby increasing the CPU requirements if you plan on simultaneously
encoding multiple programs.

NOTE As of 2003-04-15 the current IVTV driver does not support multiple
hardware encoder cards in the same chassis.

Here are a few data points

A PIII/733Mhz system can encode one video stream using the MPEG-4
codec using 480x480 capture resolution. This does not allow for live TV
watching, but does allow for encoding video and then watching it later.
The developer states that his AMD1800+ system can almost encode
two MPEG4 video streams and watch one program simultaneously.
A PIII/800Mhz system with 512MB RAM can encode one video stream
using the RTJPEG codec with 480x480 capture resolution and play it back
simultaneously, thereby allowing live TV watching.
A dual Celeron/450Mhz is able to view a 480x480 MPEG4/3300Kbps
file created on a different system with 30% CPU usage.
A P4 2.4Ghz machine can encode two 3300Kbps 480x480 MPEG4 files
and simultaneously serve content to a remote frontend.


3.3.2 Memory

A MythTV host that is both a backend and a frontend, and that uses software
encoding with a single capture card, should run comfortably in 256 MB of
RAM; depending on other hardware choices, even less (128 MB) might suffice.
Any common type of RAM in use today is fast enough. Additional RAM is
useful, but it mainly serves as buffer space to smooth the process of
sync'ing to the hard disk. For that reason, a swap partition is effectively
useless.


3.3.3 Hard Disk(s)

Encoded video takes up a lot of hard disk space. The exact amount depends
on the encoding scheme, the size of the raw images, and the frames per
second, but a typical value for MythTV is 2 GB/hour. Allow enough space.

Because a lot of data is being written to disk, the disk used for video
storage must support DMA transfers. This is not just a disk issue. The
Linux kernel must be set tp use DMA, and the motherboard's IDE chipset must
support it. These are kernel-level issues, so all we will say here is that
you need to be sure that your kernel will support DMA and your
motherboard's chipset (look in the "ATA/IDE/MFM/RLL support" section in
"make menuconfig" for more information).

If DMA works but is not active in the kernel ... these days, most Linux
distributions turn it off by default ... you can change that from the
command line or during the boot/init sequence. You do this with the command
"hdparm". As root, or from a suitable init script, run it as follows

hdparm -d 1 /dev/hdc

replacing "hdc" with the appropriate device entry for your capture drive
(the drive, not the partition). For more on this, see the Troubleshooting
Section of this HowTo.


3.3.4 Video Capture Card

The system needs one of more video-capture cards for which Linux kernel
modules exist. We know of no complete list of video-capture cards known to
work with Linux; the "Cards" file that ships with kernel-bttv documentation
is one place to check. The most common, inexpensive cards available are
cards that use the bt848 or bt878 vidcap chip; examples are the "Hauppauge
WinTV Go" card and the "AverTV Desktop PVR" card. They use the bttv kernel
module.

After you have installed a suitable card in a pci slot, you can check that
the kernel sees it with "lspci". Look for an entry labeled "Multimedia
video controller". To get more detailed information about the card, use
"lspci -v" or "lspci -vv".

While inexpensive video-capture cards just capture raw frames, leaving
encoding to software, some higher-end cards incorporate hardware-level
encoding. Using either a G200 MJPEG encoder card, or a WinTV PVR-250 or 350
from Hauppauge and the driver from the IvyTV project
http//ivtv.sourceforge.net/ will allow you to use dedicated hardware
encoders rather than your CPU. As of 2003-04-15, the CVS version MythTV is
now able to use the PVR-250/350 cards as an input device for live TV and
for scheduled recordings. Seek support is still being written. Using the
onboard MPEG-2 encoder drastically reduces the CPU requirements for
encoding. Here are some data points

An Athlon 1800XP can decode a 720x480 8Mbps MPEG-2 file using 10% CPU
An Athlon 1Ghz can decode a 720x480 16Mbps MPEG-2 file using
30-50% CPU, can decode a 480x480 16Mbps
MPEG-2 using 30% CPU and approximately 30% for Live TV at 416x480.
<This section needs more work, but I lack the required expertise.>


3.3.5 Sound card

The system needs a sound card, or an onboard equivalent on the motherboard,
to play back sound and, in most cases, to record sound. Any sound card that
can be operated by the alsa (Advanced Linux Sound Architecture) kernel
modules will work with MythTV.

The usual practice for recording the audio associated with video captures
is to run a jumper cable from an audio output on the video-capture card to
the Line input on the sound card. Some video-capture cards use internal
audio tuners that work with the Linux kernel's btaudio module, but a
definitive list of these cards is hard to come by (in part because card
makers often change tuners without changing card model designations, a
practice called "slipstreaming"). See section 19.8 of this HowTo for more
information on btaudio.


3.3.6 Video Display Card

If you want to view television on a computer monitor, then you can use any
video card for which there is an XFree86 driver that supports xVideo (xv)
extensions. Check the XFree86 documentation for details if you are
uncertain about your preferred card.

Most people, though, want to view television on a television set. For this,
you need either (a) one of the relatively small number of cards with TV-out
outputs that are supported by Linux and XFree86, or (b) an external
VGA-to-TV converter.

For choice (a), MythTV users on the mythtv-users mailing list report the
following experience:

nVidia: recent nVidia cards (such as the GeForce4 MX440) are reported to
work with the nvtv drivers, available at
http//sourceforge.net/projects/nv-tv-out/; and with the drivers provided
by nVidia itself at http://www.nvidia.com/content/drivers/drivers.asp .

Matrox: recent Matrox cards (the G450 and G550 for sure, and possibly
others) are reported to work with kernel patches (for Linux kernel 2.4.19)
available at http//www.bglug.ca/matrox_tvout/. The URL provides a detailed
recipe for geting these cards to work (follow it *exactly*).

Savage: Savage cards with TV-out are fully supported by the standard
XFree86 driver, the latest version of which is available at
http//www.probo.com/timr/savage40.html. Some cards may require use of the
"s3switch utility" also available at that URL. NOTE: The Savage 2000 card s
unsuited for use with MythTV, but for another reason (lack of xVIdeo support).

ATI: ATI is adament that it does not support use of its cards on Linux
systems, and official XFree86 offers no support either. Some people report
making some ATI cards work with the *experimental* version of the GATOS
drivers, available in CVS at http//gatos.sourceforge.net/watching_tv.php.

These are the only four manufacturers for whom we have any information on
TV out. If anyone has successfully used other TV-out cards with Linux,
XFree86, and MythTV, please post a report on mythtv-users and we will add
that information to this HowTo.

For choice (b), MythTV users on the mythtv-users mailing list report being
satisfied with these devices:

"Web Cable Plus" from AITech.

"Averkey lite", powered by a USB port, has Composite, SVideo, YPbPr
outputs; pan, brightness, overscan/underscan controls; supports up to
1024x768 outputs; and supports PAL and NTSC.

We will add other choices as they are reported on the mythtv-users list.

--------------------DRAFT ENDS------------------------
RE: Draft revisions to HowTo section 3.3, second try [ In reply to ]
Just a comment on system requirements, in section 3.3.3 would it make sense
to discuss harddrive speeds and their relevance to system performance? I've
had great success recording a show at 480x480 MPEG-4 while watching a
previously recorded show using an Athlon 650 with 256MB PC100 SDRAM. I
believe the speed of my IDE channel and harddrive (UATA100, 7200RPM) has a
lot to do with that. I have recently put in a second tv tuner, reduced the
quality to 320x480 to compensate and I have been successful recording two
shows at once.

Just thinking out loud...

Sherm <><
--------------------------------
Kelly Reed Schuerman
kschuerman@thekeyboardcowboy.com
http://www.thekeyboardcowboy.com

-----Original Message-----
From: mythtv-users-bounces@snowman.net
[mailto:mythtv-users-bounces@snowman.net] On Behalf Of Ray Olszewski
Sent: Wednesday, April 23, 2003 1:45 PM
To: Discussion about mythtv
Subject: [mythtv-users] Draft revisions to HowTo section 3.3, second try


Based on this morning's feedback -- both directly to my first draft and on
other issues that were discussed in the last day -- I've revised the draft
update a good bit. Here is the new version, for comments and criticisms.

Thank you to everyone who offered suggestions. I invite additional
feedback, although my incorporating them into a third draft will probably
wait until I get some feedback from Robert on the baseline merits of doing
this revision.


--------------------DRAFT BEGINS---------------------

3.3 Hardware

Hardware selection is a complex topic, one this HowTo can touch on only
briefly and in general terms. The following subsections offer some general
guidence but stop short of offering specific recommendations.

For more detail about actual configurations that others have used, Mark
Cooper has setup a hardware database at
http//www.goldfish.org/~mcooper/pvrhw/. The website will let you browse
what other users have reported as their hardware configuration, and how
happy they are with the results.

To find out more about the suitability of specific hardware choices, you
can also consult the archives of the mythtv-users mailing list, or post a
question on that list.


3.3.1 CPU Type and Speed

Selection of CPU type and speed is the trickiest element of hardware
selection, mainly because there are so many tradeoffs one can make, among
number of things the MythTV device can do simultaneously, capture size, and
encoding quality.

MythTV has two modes of operation. It can function as a software video
encoder, which means that it uses a fairly generic "dumb" video capture
card to get frames of video, encodes them using the CPU on your motherboard
and writes them to disk. High-end video capture cards and devices like the
Tivo and RePlay have dedicated encoder chips which use specialized hardware
to convert the video stream to the MPEG-2 format without using the
motherboard CPU. The main CPU has the responsibility of running the
Operating System and reading and writing the encoded frames to the disk.
These tasks have fairly low CPU requirements compared to encoding video,
which is why a device like a Series 1 Tivo can run with only 16MB of RAM
and a 54Mhz CPU.

There are many variables that go into the question "How fast a CPU do I
need to run MythTV"? Obviously, the faster your CPU, the better your
experience will be with MythTV. If you are using the software MPEG-4
encoder and performing the "Watch TV" function, where the CPU is both
encoding and decoding video simultaneously to allow Pause, Fast Forward and
Rewind functions for live TV requires more CPU then just encoding or
decoding. MythTV also supports multiple encoder cards in a single PC,
thereby increasing the CPU requirements if you plan on simultaneously
encoding multiple programs.

NOTE As of 2003-04-15 the current IVTV driver does not support multiple
hardware encoder cards in the same chassis.

Here are a few data points

A PIII/733Mhz system can encode one video stream using the MPEG-4
codec using 480x480 capture resolution. This does not allow for live TV
watching, but does allow for encoding video and then watching it later.
The developer states that his AMD1800+ system can almost encode
two MPEG4 video streams and watch one program simultaneously.
A PIII/800Mhz system with 512MB RAM can encode one video stream
using the RTJPEG codec with 480x480 capture resolution and play it back
simultaneously, thereby allowing live TV watching.
A dual Celeron/450Mhz is able to view a 480x480 MPEG4/3300Kbps
file created on a different system with 30% CPU usage.
A P4 2.4Ghz machine can encode two 3300Kbps 480x480 MPEG4 files
and simultaneously serve content to a remote frontend.


3.3.2 Memory

A MythTV host that is both a backend and a frontend, and that uses software
encoding with a single capture card, should run comfortably in 256 MB of
RAM; depending on other hardware choices, even less (128 MB) might suffice.
Any common type of RAM in use today is fast enough. Additional RAM is
useful, but it mainly serves as buffer space to smooth the process of
sync'ing to the hard disk. For that reason, a swap partition is effectively
useless.


3.3.3 Hard Disk(s)

Encoded video takes up a lot of hard disk space. The exact amount depends
on the encoding scheme, the size of the raw images, and the frames per
second, but a typical value for MythTV is 2 GB/hour. Allow enough space.

Because a lot of data is being written to disk, the disk used for video
storage must support DMA transfers. This is not just a disk issue. The
Linux kernel must be set tp use DMA, and the motherboard's IDE chipset must
support it. These are kernel-level issues, so all we will say here is that
you need to be sure that your kernel will support DMA and your
motherboard's chipset (look in the "ATA/IDE/MFM/RLL support" section in
"make menuconfig" for more information).

If DMA works but is not active in the kernel ... these days, most Linux
distributions turn it off by default ... you can change that from the
command line or during the boot/init sequence. You do this with the command
"hdparm". As root, or from a suitable init script, run it as follows

hdparm -d 1 /dev/hdc

replacing "hdc" with the appropriate device entry for your capture drive
(the drive, not the partition). For more on this, see the Troubleshooting
Section of this HowTo.


3.3.4 Video Capture Card

The system needs one of more video-capture cards for which Linux kernel
modules exist. We know of no complete list of video-capture cards known to
work with Linux; the "Cards" file that ships with kernel-bttv documentation
is one place to check. The most common, inexpensive cards available are
cards that use the bt848 or bt878 vidcap chip; examples are the "Hauppauge
WinTV Go" card and the "AverTV Desktop PVR" card. They use the bttv kernel
module.

After you have installed a suitable card in a pci slot, you can check that
the kernel sees it with "lspci". Look for an entry labeled "Multimedia
video controller". To get more detailed information about the card, use
"lspci -v" or "lspci -vv".

While inexpensive video-capture cards just capture raw frames, leaving
encoding to software, some higher-end cards incorporate hardware-level
encoding. Using either a G200 MJPEG encoder card, or a WinTV PVR-250 or 350
from Hauppauge and the driver from the IvyTV project
http//ivtv.sourceforge.net/ will allow you to use dedicated hardware
encoders rather than your CPU. As of 2003-04-15, the CVS version MythTV is
now able to use the PVR-250/350 cards as an input device for live TV and
for scheduled recordings. Seek support is still being written. Using the
onboard MPEG-2 encoder drastically reduces the CPU requirements for
encoding. Here are some data points

An Athlon 1800XP can decode a 720x480 8Mbps MPEG-2 file using 10%
CPU
An Athlon 1Ghz can decode a 720x480 16Mbps MPEG-2 file using
30-50% CPU, can decode a 480x480 16Mbps
MPEG-2 using 30% CPU and approximately 30% for Live TV at 416x480.
<This section needs more work, but I lack the required expertise.>


3.3.5 Sound card

The system needs a sound card, or an onboard equivalent on the motherboard,
to play back sound and, in most cases, to record sound. Any sound card that
can be operated by the alsa (Advanced Linux Sound Architecture) kernel
modules will work with MythTV.

The usual practice for recording the audio associated with video captures
is to run a jumper cable from an audio output on the video-capture card to
the Line input on the sound card. Some video-capture cards use internal
audio tuners that work with the Linux kernel's btaudio module, but a
definitive list of these cards is hard to come by (in part because card
makers often change tuners without changing card model designations, a
practice called "slipstreaming"). See section 19.8 of this HowTo for more
information on btaudio.


3.3.6 Video Display Card

If you want to view television on a computer monitor, then you can use any
video card for which there is an XFree86 driver that supports xVideo (xv)
extensions. Check the XFree86 documentation for details if you are
uncertain about your preferred card.

Most people, though, want to view television on a television set. For this,
you need either (a) one of the relatively small number of cards with TV-out
outputs that are supported by Linux and XFree86, or (b) an external
VGA-to-TV converter.

For choice (a), MythTV users on the mythtv-users mailing list report the
following experience:

nVidia: recent nVidia cards (such as the GeForce4 MX440) are
reported to
work with the nvtv drivers, available at
http//sourceforge.net/projects/nv-tv-out/; and with the drivers provided
by nVidia itself at http://www.nvidia.com/content/drivers/drivers.asp .

Matrox: recent Matrox cards (the G450 and G550 for sure, and
possibly
others) are reported to work with kernel patches (for Linux kernel 2.4.19)
available at http//www.bglug.ca/matrox_tvout/. The URL provides a detailed
recipe for geting these cards to work (follow it *exactly*).

Savage: Savage cards with TV-out are fully supported by the standard

XFree86 driver, the latest version of which is available at
http//www.probo.com/timr/savage40.html. Some cards may require use of the
"s3switch utility" also available at that URL. NOTE: The Savage 2000 card s
unsuited for use with MythTV, but for another reason (lack of xVIdeo
support).

ATI: ATI is adament that it does not support use of its cards on
Linux
systems, and official XFree86 offers no support either. Some people report
making some ATI cards work with the *experimental* version of the GATOS
drivers, available in CVS at http//gatos.sourceforge.net/watching_tv.php.

These are the only four manufacturers for whom we have any information on
TV out. If anyone has successfully used other TV-out cards with Linux,
XFree86, and MythTV, please post a report on mythtv-users and we will add
that information to this HowTo.

For choice (b), MythTV users on the mythtv-users mailing list report being
satisfied with these devices:

"Web Cable Plus" from AITech.

"Averkey lite", powered by a USB port, has Composite, SVideo,
YPbPr
outputs; pan, brightness, overscan/underscan controls; supports up to
1024x768 outputs; and supports PAL and NTSC.

We will add other choices as they are reported on the mythtv-users list.

--------------------DRAFT ENDS------------------------

_______________________________________________
mythtv-users mailing list
mythtv-users@snowman.net
http://lists.snowman.net/cgi-bin/mailman/listinfo/mythtv-users
RE: Draft revisions to HowTo section 3.3, second try [ In reply to ]
At 02:58 PM 4/23/2003 -0500, Kelly Reed Schuerman wrote:
>Just a comment on system requirements, in section 3.3.3 would it make sense
>to discuss harddrive speeds and their relevance to system performance? I've
>had great success recording a show at 480x480 MPEG-4 while watching a
>previously recorded show using an Athlon 650 with 256MB PC100 SDRAM. I
>believe the speed of my IDE channel and harddrive (UATA100, 7200RPM) has a
>lot to do with that. I have recently put in a second tv tuner, reduced the
>quality to 320x480 to compensate and I have been successful recording two
>shows at once.
>
>Just thinking out loud...

Saying *something* about this topic appeals to me. But ... what? Aside from
the fact that DMA support is a must, what do we as a group concretely know
about performance requirements for IDE drives? (For that matter, does
anyone here use SCSI? What are those of you who report capturing to NFS
mounts use?)

What rotational speeds, seek times, UDMA levels are needed? How much does
it matter if the live-buffer location is on a different physical drive from
the long-term-storage partition? If you use 2 drives, how important is it
that they be on different IDE channels?

Because hard drives for vidcap have to be big, they typically have to be
fairly new as well. My only actual vidcap experience is with drives that
are 7200 RPM, use UDMA133, and are les than 18 months old. I don't know
whether a 5200 RPM drive, or an UDMA33 drive, would work or not, and I
hesitate to say it would not given my lack of presonal knowledge. I have
guesses about the answers to some of the other questions I posed, but
guesses aren't answers, not for a HowTo, anyway.

But is there someone else here who knows something more concrete than I, or
than Kelly's imprecise "I believe ..."?
Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
Ray Olszewski wrote:
> Based on this morning's feedback -- both directly to my first draft and
> on other issues that were discussed in the last day -- I've revised the
> draft update a good bit. Here is the new version, for comments and
> criticisms.

Here are a few suggestions and comments in the spirit of
improving the documentation. You've taken on some touchy
issues and handled them well. I especially like your
explanation of DMA.

...
> 3.3.1 CPU Type and Speed
>
> Selection of CPU type and speed is the trickiest element of hardware
> selection, mainly because there are so many tradeoffs one can make,
> among number of things the MythTV device can do simultaneously, capture
> size, and encoding quality.
>
> MythTV has two modes of operation. It can function as a software video
> encoder, which means that it uses a fairly generic "dumb" video capture
> card to get frames of video, encodes them using the CPU on your
> motherboard and writes them to disk. High-end video capture cards and
> devices like the Tivo and RePlay have dedicated encoder chips which use
ReplayTV

...
> 3.3.3 Hard Disk(s)
>
> Encoded video takes up a lot of hard disk space. The exact amount
> depends on the encoding scheme, the size of the raw images, and the
> frames per second, but a typical value for MythTV is 2 GB/hour. Allow
> enough space.

typical values for MythTV range from 700MB/hour to 2 GB/hour.

[the default mpeg4 parameters are less than a 800MB/hr.]

...
> 3.3.4 Video Capture Card
>
> The system needs one of more video-capture cards for which Linux kernel
> modules exist. We know of no complete list of video-capture cards known
> to work with Linux; the "Cards" file that ships with kernel-bttv
> documentation is one place to check. The most common, inexpensive cards
> available are cards that use the bt848 or bt878 vidcap chip; examples
> are the "Hauppauge WinTV Go" card and the "AverTV Desktop PVR" card.
> They use the bttv kernel module.
>
> After you have installed a suitable card in a pci slot, you can check
> that the kernel sees it with "lspci". Look for an entry labeled
> "Multimedia video controller". To get more detailed information about
> the card, use "lspci -v" or "lspci -vv".

[.this paragraph is installation/configuration and probably
belongs in a video troubleshooting section (which would need
to cover more than lspci ;-). The current docs seem to assume
that not only do you have a card installed, the driver is
loaded and you can probe and read video from /dev/video.]

...
> 3.3.5 Sound card
>
> The system needs a sound card, or an onboard equivalent on the
> motherboard, to play back sound and, in most cases, to record sound. Any
> sound card that can be operated by the alsa (Advanced Linux Sound
> Architecture) kernel modules will work with MythTV.

[.Is this true? Might there be cards that have ALSA support
that do not handle full duplex properly?]

...
> 3.3.6 Video Display Card
>
> If you want to view television on a computer monitor, then you can use
> any video card for which there is an XFree86 driver that supports xVideo
> (xv) extensions. Check the XFree86 documentation for details if you are
> uncertain about your preferred card.
>
> Most people, though, want to view television on a television set. For
> this, you need either (a) one of the relatively small number of cards
> with TV-out outputs that are supported by Linux and XFree86, or (b) an
> external VGA-to-TV converter.
>
> For choice (a), MythTV users on the mythtv-users mailing list report the
> following experience:

[Replace the following paragraph]
>
> nVidia: recent nVidia cards (such as the GeForce4 MX440) are
> reported to work with the nvtv drivers, available at
> http//sourceforge.net/projects/nv-tv-out/; and with the drivers
> provided by nVidia itself at
> http://www.nvidia.com/content/drivers/drivers.asp .
>

nVidia: all cards with tv-out that are supported by
the current drivers from nVidia should be able to display
on a television. Some cards with certain chip sets support
overscan adjustment.

See: http://www.nvidia.com/content/drivers/drivers.asp

Many older nVidia cards (such as GeForce2 cards with tv-out)
can use <A HREF=http//sourceforge.net/projects/nv-tv-out/>nvtv</a>,
a utility which provides controls for overscan, x,y position
and several other useful controls to fine tune output.

[.GeForce4 MX440 is a card that specifically does not work
with nvtv ;-]

...
> ATI: ATI is adament that it does not support use of its cards on
> Linux systems, and official XFree86 offers no support either. Some
> people report making some ATI cards work with the *experimental* version

with the *experimental* "devel" branch

> of the GATOS drivers, available in CVS at
> http//gatos.sourceforge.net/watching_tv.php.
...
Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
Ray Olszewski wrote:
> Based on this morning's feedback -- both directly to my first draft and
> on other issues that were discussed in the last day -- I've revised the
> draft update a good bit. Here is the new version, for comments and
> criticisms.

Opps, too much trimming of my last message. I cut out one
of the comments.

...

> 3.3.2 Memory
>
> A MythTV host that is both a backend and a frontend, and that uses
> software encoding with a single capture card, should run comfortably in
> 256 MB of RAM; depending on other hardware choices, even less (128 MB)
> might suffice. Any common type of RAM in use today is fast enough.
> Additional RAM is useful, but it mainly serves as buffer space to smooth
> the process of sync'ing to the hard disk. For that reason, a swap
> partition is effectively useless.

[.The last sentence is unnecessary and misleading. This
shouldn't imply that you should not have a swap partition.
If anything, maybe:]

You will want to have enough RAM so that memory paging
will not impact your system performance.

-- bjm
Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
Ray Olszewski wrote:

> I don't know whether a 5200 RPM drive, or an UDMA33 drive, would
> work or not, and I hesitate to say it would not given my lack of
> presonal knowledge.

I have 2 Seagate 80 GB 5200 disks here happily working on a single IDE
channel. I specifically baught 5200 disks for the reduced heat and
hopefully better reliability.

When I copied a huge directory (18 GB), it took extremely long (1.5
hours?), and I don't know why, yet recording while playback (on another
machine) works relatively well, together with an Athlon 1200 (I have
occasional framedrops at panning cameras, though, not sure why).



Uniform Multi-Platform E-IDE driver Revision: 6.31
ide: Assuming 33MHz system bus speed for PIO modes; override with idebus=xx
SIS5513: IDE controller on PCI bus 00 dev 15
SIS5513: detected chipset, but driver not compiled in!
SIS5513: chipset revision 208
SIS5513: not 100% native mode: will probe irqs later
ide0: BM-DMA at 0xff00-0xff07, BIOS settings: hda:DMA, hdb:DMA
ide1: BM-DMA at 0xff08-0xff0f, BIOS settings: hdc:DMA, hdd:DMA
hda: ST380020A, ATA DISK drive
hdb: ST380020A, ATA DISK drive
hdc: LITEON DVD-ROM LTD163D, ATAPI CD/DVD-ROM drive
ide0 at 0x1f0-0x1f7,0x3f6 on irq 14
ide1 at 0x170-0x177,0x376 on irq 15
hda: 156301488 sectors (80026 MB) w/2048KiB Cache, CHS=9729/255/63
hdb: 156301488 sectors (80026 MB) w/2048KiB Cache, CHS=9729/255/63


# hdparm -I /dev/hda

/dev/hda:

ATA device, with non-removable media
Model Number: ST380020A
Serial Number: xxxxx
Firmware Revision: xxx
Standards:
Supported: 6 5 4 3
Likely used: 6
Configuration:
Logical max current
cylinders 16383 4047
heads 16 16
sectors/track 63 255
--
CHS current addressable sectors: 16511760
LBA user addressable sectors: 156301488
device size with M = 1024*1024: 76319 MBytes
device size with M = 1000*1000: 80026 MBytes (80 GB)
Capabilities:
LBA, IORDY(can be disabled)
bytes avail on r/w long: 4 Queue depth: 1
Standby timer values: spec'd by Standard
R/W multiple sector transfer: Max = 16 Current = 16
Advanced power management level: unknown setting (0x0040)
Recommended acoustic management value: 128, current value: 254
DMA: mdma0 mdma1 mdma2 udma0 udma1 udma2 udma3 udma4 *udma5
Cycle time: min=120ns recommended=120ns
PIO: pio0 pio1 pio2 pio3 pio4
Cycle time: no flow control=240ns IORDY flow control=120ns
Commands/features:
Enabled Supported:
* READ BUFFER cmd
* WRITE BUFFER cmd
* Host Protected Area feature set
* Look-ahead
* Write cache
* Power Management feature set
Security Mode feature set
* SMART feature set
* Device Configuration Overlay feature set
* Automatic Acoustic Management feature set
SET MAX security extension
* Advanced Power Management feature set
* DOWNLOAD MICROCODE cmd
* SMART self-test
* SMART error logging
Security:
Master password revision code = 65534
supported
not enabled
not locked
not frozen
not expired: security count
not supported: enhanced erase
HW reset results:
CBLID- above Vih
Device num = 0 determined by the jumper
Checksum: correct
Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
Ray Olszewski wrote:
> At 02:58 PM 4/23/2003 -0500, Kelly Reed Schuerman wrote:
>
>> Just a comment on system requirements, in section 3.3.3 would it make
>> sense
>> to discuss harddrive speeds and their relevance to system
>> performance? I've
>> had great success recording a show at 480x480 MPEG-4 while watching a
>> previously recorded show using an Athlon 650 with 256MB PC100 SDRAM. I
>> believe the speed of my IDE channel and harddrive (UATA100, 7200RPM)
>> has a
>> lot to do with that. I have recently put in a second tv tuner, reduced
>> the
>> quality to 320x480 to compensate and I have been successful recording two
>> shows at once.
>>
>> Just thinking out loud...
>
>
> Saying *something* about this topic appeals to me. But ... what? Aside
> from the fact that DMA support is a must, what do we as a group
> concretely know about performance requirements for IDE drives? (For that
> matter, does anyone here use SCSI? What are those of you who report
> capturing to NFS mounts use?)
>
> What rotational speeds, seek times, UDMA levels are needed? How much
> does it matter if the live-buffer location is on a different physical
> drive from the long-term-storage partition? If you use 2 drives, how
> important is it that they be on different IDE channels?
>
> Because hard drives for vidcap have to be big, they typically have to be
> fairly new as well. My only actual vidcap experience is with drives that
> are 7200 RPM, use UDMA133, and are les than 18 months old. I don't know
> whether a 5200 RPM drive, or an UDMA33 drive, would work or not, and I
> hesitate to say it would not given my lack of presonal knowledge. I have
> guesses about the answers to some of the other questions I posed, but
> guesses aren't answers, not for a HowTo, anyway.

I agree and I see this is often misunderstood. Realize that
there are 3600sec/hr so a 3.6GB/hr output file writes 1MB/sec.
Most myth users are writing less than 0.5MB/sec and in Kelly's
case, less than 200KB/sec.

Disk throughput doesn't matter. It is only the efficiency of
DMA and CPU time for IDE that will have an impact. However,
I have no idea of how to compare or benchmark these factors.
I do know that TiVo and ReplayTV hackers seek out 5400RPM disks
because they can easily write fast enough and they hope these
will run quieter and cooler.

-- bjm
Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
Bruce Markey wrote:

> Ray Olszewski wrote:
>
>> For that reason, a swap partition is effectively useless.
>
> [.The last sentence is unnecessary and misleading. This
> shouldn't imply that you should not have a swap partition.
> If anything, maybe:]
>
> You will want to have enough RAM so that memory paging
> will not impact your system performance.

Personally, I really prefer not to have any swap partition. Swap is a
total perf killer (given today's RAM prices) and only hides problems.
I'd rather have a mythfilldatabase process die because of too little RAM
(this did happen here with 128 MB) than to have it cause swapping, which
might bring down a recording in process.
RE: Draft revisions to HowTo section 3.3, second try [ In reply to ]
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Ray, I hadn't posted a response to your original draft because I was
incorporating large chunks of it into the HOWTO. For all "second
draft" contributors, please take a look at the document that I've
uploaded to CVS and use that as the baseline for further comments, or
to correct me where I was totally off base.

Thanks.

-----BEGIN PGP SIGNATURE-----
Version: PGPfreeware 7.0.3 for non-commercial use <http://www.pgp.com>

iQA/AwUBPqcXp/c1NpCTlP0JEQIZgQCgrdcPfYcisVOzsmgk5P+HR4woE1wAoNgb
u1tWxSe999/JfcNE9Imu/OVo
=8u7d
-----END PGP SIGNATURE-----
RE: Draft revisions to HowTo section 3.3, second try [ In reply to ]
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

> I don't know whether a 5200 RPM drive, or an UDMA33 drive, would
> work or not, and I hesitate to say it would not given my lack of
> presonal knowledge.

I know that they work; I'm using some. Remember, if you're using
MPEG-4 @ the default of 2200Kbps, (or even higher values), you're
still within the typical performance range of older drives. 2Mbps
isn't a whole lot.

-----BEGIN PGP SIGNATURE-----
Version: PGPfreeware 7.0.3 for non-commercial use <http://www.pgp.com>

iQA/AwUBPqcYUvc1NpCTlP0JEQI/qQCg0govV7YcpnC6ZobsW5sd0zO07ZwAoO55
nLH6lrbyHYEl2E00nfYpCA3R
=mUVj
-----END PGP SIGNATURE-----
Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
Ben Bucksch wrote:

> ide: Assuming 33MHz system bus speed for PIO modes; override with
> idebus=xx
> SIS5513: detected chipset, but driver not compiled in!

# hdparm -t /dev/hda
Timing buffered disk reads: 64 MB in 13.76 seconds = 4.65 MB/sec
# hdparm -t /dev/hdb
Timing buffered disk reads: 64 MB in 13.80 seconds = 4.64 MB/sec

No good (actually in line with that huge, slow copy) (my desktop has
35-46 MB/s). Yet MythTV works. Meaning that requirements to disks seem
to be fairly low.

I didn't realise that the IDE driver for my chipset was not compiled
into the kernel on that machine (*cough*). I did that now and now have
normal disk speeds.

You could include a hint to
hdparm -I /dev/hda
hdparm -t /dev/hda
dmesg
Specifically, if hdparm -t says something above 20 or 30 MB/s, all
should be fine, no need to worry. Maybe say that right in the beginning.



ide: Assuming 33MHz system bus speed for PIO modes; override with idebus=xx
hda: 156301488 sectors (80026 MB) w/2048KiB Cache, CHS=9729/255/63,
UDMA(100)
hdb: 156301488 sectors (80026 MB) w/2048KiB Cache, CHS=9729/255/63,
UDMA(100)

# hdparm -t /dev/hda
Timing buffered disk reads: 64 MB in 2.30 seconds = 27.83 MB/sec
# hdparm -t /dev/hdb
Timing buffered disk reads: 64 MB in 2.63 seconds = 24.33 MB/sec
RE: Draft revisions to HowTo section 3.3, second try [ In reply to ]
At 05:46 PM 4/23/2003 -0500, Robert Kulagowski wrote:
>
>-----BEGIN PGP SIGNED MESSAGE-----
>Hash: SHA1
>
>Ray, I hadn't posted a response to your original draft because I was
>incorporating large chunks of it into the HOWTO. For all "second
>draft" contributors, please take a look at the document that I've
>uploaded to CVS and use that as the baseline for further comments, or
>to correct me where I was totally off base.

Reading the raw HTML that comes out of CVS is a bit of a chore, so please
forgive me if these comments reflect a misreading of the changes you made.

First, I would restore the comment I made about swap partitions being
useless for MythTV. Of course, this is just my opinion, and you might not
share it.

Second, I would restore the comment I made that additional RAM (over 256
MB) acts to buffer writes to the hard disk. This role of "spare" RAM is
commonly overlooked, and people should be more aware of it. On any system
that runs a real-time, use-it-or-lose-it activity like video capture, the
extra time buffering buys can help the system cope with transient encoding
loads (camera pans seems now to be everyone's main example of this, for
good reason).

Third, the section on Savage video cards seems not to have made it in. I'd
recommend restoring that; the many people who corrected my original version
are surely right about the usability of those cards, but documentation ...
not just yours; most video apps ... seems never to mention them.

Other than that ... I'll infer from your actions that you find this
material useful, and start work on a similar revision to the next section
on software requirements.
Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
On Wednesday 23 April 2003 07:57 pm, Ray Olszewski wrote:
> First, I would restore the comment I made about swap partitions being
> useless for MythTV. Of course, this is just my opinion, and you might not
> share it.

The linux kernel is tuned to run better with, if not essentially require, some
swap space.

> Second, I would restore the comment I made that additional RAM (over 256
> MB) acts to buffer writes to the hard disk. This role of "spare" RAM is
> commonly overlooked, and people should be more aware of it. On any system
> that runs a real-time, use-it-or-lose-it activity like video capture, the
> extra time buffering buys can help the system cope with transient encoding
> loads (camera pans seems now to be everyone's main example of this, for
> good reason).

MythTV syncs all writes to disk approximately once a second. Otherwise, the
kernel will buffer lots of data, then decide to write out _lots_ of it at
once, basically killing performance and causing frames to be dropped.

Isaac
Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
Ray Olszewski wrote:
...
> First, I would restore the comment I made about swap partitions being
> useless for MythTV. Of course, this is just my opinion, and you might
> not share it.

Ray, I can't even parse the sentence from your draft into
anything meaningful much less useful information for
configuring a system. Swap partitions are useful and a key
element in a virtual memory system.

If you meant "don't mount a swap partition", you didn't say
that. If so, it is not the place of any application
documentation to tell anyone using any virtual memory O/S
that they should not have paging space. This is so obviously
wrong that I doubt that is what you meant.

If you meant "you will want to have enough RAM so that memory
paging will not impact your system performance", your comment
didn't express that either. This, or some similar point, I
would agree with. However, simply saying swap partition are
"useless" in the documentation doesn't tell users what they
need to do or what they need to know.

-- bjm
Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
Ben Bucksch wrote:
> Bruce Markey wrote:
>
>> Ray Olszewski wrote:
>>
>>> For that reason, a swap partition is effectively useless.
>>
>>
>> [.The last sentence is unnecessary and misleading. This
>> shouldn't imply that you should not have a swap partition.
>> If anything, maybe:]
>>
>> You will want to have enough RAM so that memory paging
>> will not impact your system performance.
>
>
> Personally, I really prefer not to have any swap partition.

Great, but let's first establish that this is off-topic. No
application documentation should tell anyone that a Linux
system should not have a swap partition. Ray's comment doesn't
say that but it seems misleading.

> Swap is a
> total perf killer (given today's RAM prices) and only hides problems.

Swap is a performance win that alleviates a lot of problems that
would otherwise be fatal. It's the lack of swap in your example
that 'hides' the problem and solution.

> I'd rather have a mythfilldatabase process die because of too little RAM
> (this did happen here with 128 MB) than to have it cause swapping, which
> might bring down a recording in process.

So, you deliberately allow malloc() to fail as a preventative
measure against possible page-outs? Not a good idea. When
you do run out of virtual memory, what's to say that it isn't
mythbackend that next fails to malloc and is the process that
dies (or cron or syslogd or inetd)? Running out of virtual
memory is a nasty thing whether or not swap space is involved.
Relying on malloc failures (BTW lots of programs don't check
the return status of malloc()) isn't what you want.

When you have swap space, as active processes need more memory,
pages from inactive processes can be moved out and left out
until they become active again. This is normally a one time
thing. The behavior you are worried about is when the active
processes need more memory than the amount of physical memory.
In that case, active pages need to be constantly moved in and
out. The solution to this thrashing is to buy more memory
regardless of how much swap space you have.

The problem with not having a swap partition is that there is
no place to put memory allocated to inactive processes.
Therefore you shrink the amount of memory available for the
active processes which exasperates your memory shortage.

Not having a swap partition may seem like a sure fire way to
prevent thrashing but it doesn't fix anything, it only makes
your memory usage less efficient and increases the chances of
crashes and mis-behavior.

-- bjm
Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
> 3.3.2 Memory
>
> A MythTV host that is both a backend and a frontend, and that uses
> software encoding with a single capture card, should run comfortably in
> 256 MB of RAM; depending on other hardware choices, even less (128 MB)
> might suffice. Any common type of RAM in use today is fast enough.
> Additional RAM is useful, but it mainly serves as buffer space to smooth
> the process of sync'ing to the hard disk. For that reason, a swap
> partition is effectively useless.

My experience relates directly to this issue. I have experimented with two
different motherboards (Via VA6 and Shuttle FV25) two different processors
(PIII 700 MHz and Celeron 1.4 GHz), two different sets of memory SIMMs
(PC100 and PC133), and various clock speeds. While processor speed is
certainly a factor, the primary factor in MythTV performance is memory
access speed. By "MythTV performance" I mean the maximum resolution and
quality readily achievable with a given codec for recording only.

For example, the same motherboard (VA6) with the same memory SIMMs (PC100
and also PC133) performed nearly identically with a PIII-700 and
Celeron-1.4 on MythTV.

The same processor (Celeron-1.4) on the same motherboard (VA6 and also
FV25) with different SIMMs (PC100 vs PC133, with memory clock speed
adjusted accordingly) had a dramatic effect on MythTV performance.

In overclocking mode, tweaking the processor clock speed did not have
nearly the effect that tweaking the memory clock speed did.

This all makes sense: video applications typically have high miss rates at
the cache and are thus memory bound.

The conclusion: faster memory systems will work better. If you can, avoid
PC100 memory and systems. Similarily, if you can, use DDR memory systems.

Does anyone have a set up with other motherboards (eg, Via EPIA) who can
perform similar tests between memory speeds? Or a motherboard which
supports PC133 as well as DDR?

- pz.
RE: Draft revisions to HowTo section 3.3, second try [ In reply to ]
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

> Reading the raw HTML that comes out of CVS is a bit of a
> chore, so please
> forgive me if these comments reflect a misreading of the
> changes you made.

the .sgml file is the raw source, complete with markup tags, and the
.txt has the same content in it as the .html version.

> First, I would restore the comment I made about swap partitions
> being useless for MythTV. Of course, this is just my opinion, and
> you might not share it.

The reason that I removed this statement is that I concur with Bruce
on this issue. Saying that swap is useless is overly broad, and
makes assumptions about the machine / user that we can't make without
having a fixed hardware configuration.

> Second, I would restore the comment I made that additional
> RAM (over 256
> MB) acts to buffer writes to the hard disk. This role of
> "spare" RAM is
> commonly overlooked, and people should be more aware of it.
> On any system
> that runs a real-time, use-it-or-lose-it activity like video
> capture, the
> extra time buffering buys can help the system cope with
> transient encoding
> loads (camera pans seems now to be everyone's main example of
> this, for
> good reason).
>

I'll put in some wording to reflect this and what Isaac said.

> Third, the section on Savage video cards seems not to have
> made it in. I'd
> recommend restoring that; the many people who corrected my
> original version
> are surely right about the usability of those cards, but
> documentation ...
> not just yours; most video apps ... seems never to mention them.

The information was changing too rapidly at the time I was editing
the document. Now that there have been further responses, do you
want to re-write and send to the list again?

-----BEGIN PGP SIGNATURE-----
Version: PGPfreeware 7.0.3 for non-commercial use <http://www.pgp.com>

iQA/AwUBPqf1ePc1NpCTlP0JEQIZagCgsO8F+dEMVb9JBFRnIR3mfQWSngMAoKcv
fWlB0QkuI9uGWDsUNnArBfdk
=aT7E
-----END PGP SIGNATURE-----
Re: Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
Your comments here make me glad I wrote the original statement you are
responding to. There are too many details like this that we do not discuss,
perhaps because we all think the answer is obvious. But we arrive ay many
different "obvious" answers. There's an old saying -- "Things that go
without saying ... usually go better when said" -- that may apply here.

That said, I have a couple of questions about your report.

1. You say faster memory "had a dramatic effect on MythTV performance". In
what way? What was better about the PC133 systems?

2. You say that for either type of memory you tested, your test system
"performed nearly identically with a PIII-700 and Celeron-1.4 on MythTV".
Again, measured by what standard? My experience is that P-III and Celeron
chips are almost identical in their performance when doing video capture,
but that means that clock speed is everything in these comparisons. So I'd
expect your test to show the (slow) P-III running at almost twice the CPU
load as the (fast) Celeron.

3. Based on a comparison of PC100 and PC133 RAM, you conclude that "if you
can, use DDR memory systems". This seems to be extrapolating your
observations outside their range (PC100 versus PC133). Do you have any
additional reason for favoring DDR memory over PC133? Given its
considerably higher cost, the move isn't trivial, whereas these days it is
all but impossiable even to find PC100 memory (to be honest, when I wrote
what I did, I was not thinking of PC100 as a "ommon type of RAM in use
today" ... though you can hardly be criticized based on my imprecision).

At 08:47 AM 4/24/2003 -0400, you wrote:
> > 3.3.2 Memory
> >
> > A MythTV host that is both a backend and a frontend, and that uses
> > software encoding with a single capture card, should run comfortably in
> > 256 MB of RAM; depending on other hardware choices, even less (128 MB)
> > might suffice. Any common type of RAM in use today is fast enough.
> > Additional RAM is useful, but it mainly serves as buffer space to smooth
> > the process of sync'ing to the hard disk. For that reason, a swap
> > partition is effectively useless.
>
>My experience relates directly to this issue. I have experimented with two
>different motherboards (Via VA6 and Shuttle FV25) two different processors
>(PIII 700 MHz and Celeron 1.4 GHz), two different sets of memory SIMMs
>(PC100 and PC133), and various clock speeds. While processor speed is
>certainly a factor, the primary factor in MythTV performance is memory
>access speed. By "MythTV performance" I mean the maximum resolution and
>quality readily achievable with a given codec for recording only.
>
>For example, the same motherboard (VA6) with the same memory SIMMs (PC100
>and also PC133) performed nearly identically with a PIII-700 and
>Celeron-1.4 on MythTV.
>
>The same processor (Celeron-1.4) on the same motherboard (VA6 and also
>FV25) with different SIMMs (PC100 vs PC133, with memory clock speed
>adjusted accordingly) had a dramatic effect on MythTV performance.
>
>In overclocking mode, tweaking the processor clock speed did not have
>nearly the effect that tweaking the memory clock speed did.
>
>This all makes sense: video applications typically have high miss rates at
>the cache and are thus memory bound.
>
>The conclusion: faster memory systems will work better. If you can, avoid
>PC100 memory and systems. Similarily, if you can, use DDR memory systems.
>
>Does anyone have a set up with other motherboards (eg, Via EPIA) who can
>perform similar tests between memory speeds? Or a motherboard which
>supports PC133 as well as DDR?
Re: Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
On Thu, Apr 24, 2003 at 10:05:48AM -0700, Ray Olszewski wrote:
> Do you have any additional reason for favoring DDR memory over
> PC133? Given its considerably higher cost, the move isn't trivial,

For the record, DDR is now cheaper then PC133 per byte. If one is
starting from scratch in making any sort of computer, DDR is a very
compelling price per performance option.

crucial.com:
256MB DDR module - $41.99
256MB PC133 module - $59.99

E

--
Erik Hovland
mail: erik@hovland.org
web: http://hovland.org/
PGP/GPG public key available on request
Re: Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
At 12:20 PM 4/24/2003 -0700, Erik Hovland wrote:
>On Thu, Apr 24, 2003 at 10:05:48AM -0700, Ray Olszewski wrote:
> > Do you have any additional reason for favoring DDR memory over
> > PC133? Given its considerably higher cost, the move isn't trivial,
>
>For the record, DDR is now cheaper then PC133 per byte. If one is
>starting from scratch in making any sort of computer, DDR is a very
>compelling price per performance option.
>
>crucial.com:
>256MB DDR module - $41.99
>256MB PC133 module - $59.99

Actually, it depends a bit on where you shop. Yesterday's newspaper ad for
Fry's Electronics (located near me in Palo Alto, CA) had these prices:

512MB PC133 memory - $39.99
512MB DDR (PC2100) memory - $49.99

I would never base this sort of comparison on prices from one vendor
(especially not a relatively high-priced one, it would seem). But it does
seem that the price premium for DDR is dropping around here ... as recently
as a month ago, DDR cost twice as much as PC133 at Fry's ... so this
observation may be true soon.
Re: Draft revisions to HowTo section 3.3, second try [ In reply to ]
At the risk of contributing to a potentially off-topic thread, I
want to second Bruce's comments and also add that the kernel kills
/random/ processes in an out-of-memory scenario, in addition to
denying malloc()s for more memory. IIRC, the kernel has been designed
to use swap since somewhere around version 1.1-1.3. Please use swap.


On Apr 23, at 19:16, Bruce Markey encoded a 2.6K recording:
> The problem with not having a swap partition is that there is no
> place to put memory allocated to inactive processes. Therefore you
> shrink the amount of memory available for the active processes which
> exasperates your memory shortage.