Archives


- Beowulf
- Beowulf Announce
- Scyld-users
- Beowulf on Debian

DMA difficulties

Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.

Search

Steven Berukoff steveb at aei-potsdam.mpg.de
Sat Apr 13 17:37:42 PDT 2002


Hi all,

This question may be very slightly off-topic, so I apologize.

I'm in the process of setting up a network installation procedure using
PXE/DHCP/NFS/Kickstart w/ RH7.2 for about 150 dual Athlon nodes.  These
nodes use a Maxtor 6L080J4 80.0GB HDD and an ASUS A7M266-D motherboard,
among other things.  One particular note is that I don't need/want CDROMs
in these systems.

Now, a vendor provided me with a couple of test nodes basically to our
specifications, except that they included CDROMs and floppies.  To make a
longish story shorter, I wanted to make sure that the nodes work fine
without the CDROM.

So, I first looked into the BIOS.  I disabled (set to "None") Primary
Slave, Secondary Master/Slave (since my HDD is Primary Master), removed
the CDROM from the list of boot devices, and disabled the Secondary IDE
channel.  Then, I passed the kernel args "ide0=dma hdb=none" to try to
enforce the HDD to use DMA during the Kickstart installation.

Now, here is the kicker: regardless of the BIOS settings, if I have the
CDROM plugged in (power+IDE, on the secondary channel) the installation
takes ~ 5 times faster than if the thing isn't there.  This installation
includes installation of ~470 packages plus formatting the HDD.  That's
right, as long as the CDROM is plugged in, everything is peachy, but once
gone, things slow down.

I think this is a problem with the DMA settings, b/c when I pass
"ide=nodma" to the kernel, WITH the CD attached, performance is
slow.  However, I can't even force DMA to be used.

If anyone has any suggestions or similar experiences, please let me know.

Thanks a bunch!
Steve


=====
Steve Berukoff					tel: 49-331-5677233
Albert-Einstein-Institute			fax: 49-331-5677298
Am Muehlenberg 1, D14477 Golm, Germany		email:steveb at aei.mpg.de





More information about the Beowulf mailing list