wmr282@hotmail.com (w m r) writes:

> What happens is ping starts showing 'destination unreachable' after a
> few seconds while doing an FTP or using Netscape. The FTP or Netscape
> will just hang. Everything else in the OS keeps working. If I do a
> /etc/rc.d/init.d/network restart, it starts working again.

I don't know what could be causing this.

> Other plug-in cards I have are a S3-Trio64 video card (in one of the
> PCI-64bit slots) and a Qlogic ISP-1020 SCSI card (in one of the
> PCI-32bit slots) for the CDROM. The hard drive is an Segate 10GB IDE,
> plugged in as /dev/hda.

Never use a hard disk on the onboard IDE in a Miata. Miatas have a
chipset bug that causes data corruption on DMA transfers to anything
not in one of the 32-bit PCI slots. The IDE onboard IDE controller is
also a rather bad one. It's there only for the CDROM.

--
Måns Rullgård
mru@kth.se

w m r wrote:
> I'm trying to install Linux on a Digital PWS Alpha Miata. I am
> loading from Mandrake 7.1. It installed OK but when I use the
> network, it hangs under load.
>
> What happens is ping starts showing 'destination unreachable' after a
> few seconds while doing an FTP or using Netscape. The FTP or Netscape
> will just hang. Everything else in the OS keeps working. If I do a
> /etc/rc.d/init.d/network restart, it starts working again.
>
> I went to kernel.org and got the 2.4.23 kernel, same thing.
>
> I am using the built-in ethernet (21143) with the tulip.o driver. I
> also tried an i82257 board (in one of the PCI-32bit slots) with the
> eepro100.o driver, same result.
>
> Other plug-in cards I have are a S3-Trio64 video card (in one of the
> PCI-64bit slots) and a Qlogic ISP-1020 SCSI card (in one of the
> PCI-32bit slots) for the CDROM. The hard drive is an Segate 10GB IDE,
> plugged in as /dev/hda. I'm pretty sure the hardware is OK as I ran
> VMS on it for years (except with a SCSI hard drive).
>
> Mike
Have you noticed any odd resource consumption during these periods? Is
CPU utilization through the roof on any one process, or is a particular
slab cache growing more rapidly that you would normally expect?

--
/************************************************** *
*Neil Horman
*Software Engineer
*Red Hat, Inc., http://people.redhat.com/nhorman
*gpg keyid: 1024D / 0x92A74FA1
*http://www.keyserver.net
************************************************** */

mru@kth.se (Måns Rullgård) wrote in message news:...
> wmr282@hotmail.com (w m r) writes:

> > Other plug-in cards I have are a S3-Trio64 video card (in one of the
> > PCI-64bit slots) and a Qlogic ISP-1020 SCSI card (in one of the
> > PCI-32bit slots) for the CDROM. The hard drive is an Segate 10GB IDE,
> > plugged in as /dev/hda.
>
> Never use a hard disk on the onboard IDE in a Miata. Miatas have a
> chipset bug that causes data corruption on DMA transfers to anything
> not in one of the 32-bit PCI slots. The IDE onboard IDE controller is
> also a rather bad one. It's there only for the CDROM.

Hmm, I thought the bug was with the 64-bit PCI slots for DMA's that
crossed page boundaries. Your way would also give trouble with the
built-in tulip chip, but I tried a plug-in Intel board with same
result. Anyway, I pounded the ^%$^%$ out of that hard drive with VMS
for years (as a data drive) doing nightly backups.

I do have another Miata at home I could lug in to rule out hw problem,
but like I said, that system has worked with VMS for a long time. I
suppose I could also try a SCSI disk to see if the problem goes away!

Thanks for responding.

Mike

wmr282@hotmail.com (w m r) writes:

>> Never use a hard disk on the onboard IDE in a Miata. Miatas have a
>> chipset bug that causes data corruption on DMA transfers to anything
>> not in one of the 32-bit PCI slots. The IDE onboard IDE controller is
>> also a rather bad one. It's there only for the CDROM.
>
> Hmm, I thought the bug was with the 64-bit PCI slots for DMA's that
> crossed page boundaries.

The onboard chips are logically on the same bus as the 64-bit slots.
The 32-bit slots are behind a bridge, that somehow prevents the bug.

> Your way would also give trouble with the built-in tulip chip, but I
> tried a plug-in Intel board with same result.

An Ethernet frame is shorter than one page, so there's no problem
there.

--
Måns Rullgård
mru@kth.se

Mike:

There's a known bug in SRM when the onboard ethernet port is set to
"Auto-Negotiate". Try setting it to Twisted-Pair.

TomC

w m r wrote:

> I'm trying to install Linux on a Digital PWS Alpha Miata. I am
> loading from Mandrake 7.1. It installed OK but when I use the
> network, it hangs under load.
>
> What happens is ping starts showing 'destination unreachable' after a
> few seconds while doing an FTP or using Netscape. The FTP or Netscape
> will just hang. Everything else in the OS keeps working. If I do a
> /etc/rc.d/init.d/network restart, it starts working again.
>
> I went to kernel.org and got the 2.4.23 kernel, same thing.
>
> I am using the built-in ethernet (21143) with the tulip.o driver. I
> also tried an i82257 board (in one of the PCI-32bit slots) with the
> eepro100.o driver, same result.
>
> Other plug-in cards I have are a S3-Trio64 video card (in one of the
> PCI-64bit slots) and a Qlogic ISP-1020 SCSI card (in one of the
> PCI-32bit slots) for the CDROM. The hard drive is an Segate 10GB IDE,
> plugged in as /dev/hda. I'm pretty sure the hardware is OK as I ran
> VMS on it for years (except with a SCSI hard drive).
>
> Mike

mru@kth.se (Måns Rullgård) wrote in message news:...
> wmr282@hotmail.com (w m r) writes:
>
> >> Never use a hard disk on the onboard IDE in a Miata. Miatas have a
> >> chipset bug that causes data corruption on DMA transfers to anything
> >> not in one of the 32-bit PCI slots. The IDE onboard IDE controller is
> >> also a rather bad one. It's there only for the CDROM.
> >
> > Hmm, I thought the bug was with the 64-bit PCI slots for DMA's that
> > crossed page boundaries.
>
> The onboard chips are logically on the same bus as the 64-bit slots.
> The 32-bit slots are behind a bridge, that somehow prevents the bug.
>
> > Your way would also give trouble with the built-in tulip chip, but I
> > tried a plug-in Intel board with same result.
>
> An Ethernet frame is shorter than one page, so there's no problem
> there.

.... assuming they never cross page boundary.

Also hangs with a SCSI hard disk (Segate 4Gig). My CD is SCSI also.

Mike

wmr282@hotmail.com (w m r) writes:

> mru@kth.se (Måns Rullgård) wrote in message news:...
>> wmr282@hotmail.com (w m r) writes:
>>
>> >> Never use a hard disk on the onboard IDE in a Miata. Miatas have a
>> >> chipset bug that causes data corruption on DMA transfers to anything
>> >> not in one of the 32-bit PCI slots. The IDE onboard IDE controller is
>> >> also a rather bad one. It's there only for the CDROM.
>> >
>> > Hmm, I thought the bug was with the 64-bit PCI slots for DMA's that
>> > crossed page boundaries.
>>
>> The onboard chips are logically on the same bus as the 64-bit slots.
>> The 32-bit slots are behind a bridge, that somehow prevents the bug.
>>
>> > Your way would also give trouble with the built-in tulip chip, but I
>> > tried a plug-in Intel board with same result.
>>
>> An Ethernet frame is shorter than one page, so there's no problem
>> there.
>
> ... assuming they never cross page boundary.

The driver can easily assure that it doesn't.

--
Måns Rullgård
mru@kth.se