I am trying to install CentOS to a server that has a Tyan S2891 motherboard, 2 Opteron 875 Dual Cores and 2GB of RAM.
I have been able trying to install over a network but the tg3 driver will not load. I then installed from a DVD and on rebooting after the installation I cannot get the network drivers to load and the RAID controller in the PCI slot is not available.
It seems that when I do an lspci -tv the only a minimal amount of devices are seen.
[0000:00]-+-00.0 nVidia Corporation CK804 Memory Controller +-01.0 nVidia Corporation CK804 ISA Bridge +-01.1 nVidia Corporation CK804 SMBus +-02.0 nVidia Corporation CK804 USB Controller +-02.1 nVidia Corporation CK804 USB Controller +-06.0 nVidia Corporation CK804 IDE +-09.0-[0000:01]----07.0 ATI Technologies Inc Rage XL -0e.0-[0000:02]--
Has anyone had success with a combination of parts close to this?
Rik
On Thu, 1 Dec 2005, rik@harddata.com wrote:
I am trying to install CentOS to a server that has a Tyan S2891 motherboard, 2 Opteron 875 Dual Cores and 2GB of RAM.
I've got a Tyan S2882-D with two dual-core Opteron 270s running CentOS 4 just fine -- but it's got the AMD-8111 chipset, not the nVidia CK804.
} -----Original Message----- } From: centos-bounces@centos.org [mailto:centos-bounces@centos.org] On } Behalf Of Paul Heinlein } Sent: Thursday, December 01, 2005 11:05 AM } } I've got a Tyan S2882-D with two dual-core Opteron 270s running CentOS } 4 just fine -- but it's got the AMD-8111 chipset, not the nVidia } CK804. } } -- } Paul Heinlein <> heinlein@madboa.com <> www.madboa.com
other specs in that box?
is it a screamer?
benchmarks?
- rh
-- Robert Hanson - Abba Communications Computer & Internet Services (509) 624-7159 - www.abbacomm.net
Paul Heinlein wrote:
I've got a Tyan S2882-D with two dual-core Opteron 270s running CentOS 4 just fine -- but it's got the AMD-8111 chipset, not the nVidia CK804.
<anal> The S2882 actually the AMD8131 HyperTransport tunnel (dual PCI-X 1.0) with the AMD8131 HyperTransport bridge (legacy PC). </anal>
Just the AMD8111 would really suck. ;->
The S2895 uses the AMD8131 with the nForce Pro 2200+2050. The S2891 uses the AMD8131 with the nForce Pro 2200.
The nForce Pro 2200 provides the legacy PC, along with PCIe and newer features. The nForce Pro 2050 just adds additional PCIe channels and new features.
Robert roberth@abbacomm.net wrote:
other specs in that box? is it a screamer?
Anything with an AMD8131 will get you excellent server performance, because it's a dual PCI-X 1.0 channel solution.
Some HP and Sun servers use 2, and one Sun product even uses 3 (for 6 PCI-X 1.0 channels).
benchmarks?
For what? The CPUs? The PCI-X I/O? The PCIe channels? You have to give some context of what you'd use on the board. Otherwise the benchmarks will vary _widely_. ;->
If you're only going to use the PCI-X channels, then it really matters little what mainboard you use -- as long as it has an AMD8131 (or the newer AMD8132 -- which is a dual PCI-X 2.0).
On Thu, 1 Dec 2005, Robert wrote:
} -----Original Message----- } From: centos-bounces@centos.org [mailto:centos-bounces@centos.org] On } Behalf Of Paul Heinlein } Sent: Thursday, December 01, 2005 11:05 AM } } I've got a Tyan S2882-D with two dual-core Opteron 270s running } CentOS 4 just fine -- but it's got the AMD-8111 chipset, not the } nVidia CK804.
other specs in that box?
4 GB RAM, generic SATA HD. Otherwise, it's headless and is used mostly for ACL2 proofs.
is it a screamer?
benchmarks?
Our developers like it. :-)
One guy (who's leaving us to go work for the Beast in Redmond ... grumble, grumble) said that their most intense model took
* several hours on his G4 laptop * a couple hours on a dual G5 Xserve * 75 minutes on an older, single-processor Opteron * 15 minutes on the machine outlined above
Paul Heinlein wrote:
On Thu, 1 Dec 2005, rik@harddata.com wrote:
I am trying to install CentOS to a server that has a Tyan S2891 motherboard, 2 Opteron 875 Dual Cores and 2GB of RAM.
I've got a Tyan S2882-D with two dual-core Opteron 270s running CentOS 4 just fine -- but it's got the AMD-8111 chipset, not the nVidia CK804.
Since CentOS is based on stable/older code base, and Tyan S2891 is bleeding-edge-new, I wouldn't be surprised if it didn't work so well. My $0.02 ONLY, no 1st hand experience, YMMV, etc.
Paul Heinlein wrote:
On Thu, 1 Dec 2005, rik@harddata.com wrote:
I am trying to install CentOS to a server that has a Tyan S2891 motherboard, 2 Opteron 875 Dual Cores and 2GB of RAM.
I've got a Tyan S2882-D with two dual-core Opteron 270s running CentOS 4 just fine -- but it's got the AMD-8111 chipset, not the nVidia CK804.
Since CentOS is based on stable/older code base, and Tyan S2891 is bleeding-edge-new, I wouldn't be surprised if it didn't work so well. My $0.02 ONLY, no 1st hand experience, YMMV, etc.
This is what I figured so I have tested with Fedora Core 4 installs.
Interesting to find that the 2.6.11 install kernel works but the latest 2.6.14 kernel has the same symptoms. I am scratching my head and firing up bugzilla.
Rik
rik@harddata.com wrote:
This is what I figured so I have tested with Fedora Core 4 installs. Interesting to find that the 2.6.11 install kernel works but the latest 2.6.14 kernel has the same symptoms. I am scratching my head and firing up bugzilla.
It's really the issue of all those PCI busses. Enumeration of all those devices across all those busses is not excatly easy.
Again, it wasn't too long ago that 99% of systems had only 2 PCI busses, and servers were 3 or 4 maximum. Now we're talking 4-5 are standard these days on a desktop, and 6, 8 or even 10+ on a server.
Enumerating all those busses/devices can get rather tricky, and kernels are going to be conservative by default.
I had similar problems with the Tyan B2891 'Transport GT24' - see:
https://www.redhat.com/archives/nahant-list/2005-November/msg00062.html
Also make sure that the 'Installed OS' is set to 'Linux' in the BIOS.
James Pearson
On 01/12/05, rik@harddata.com rik@harddata.com wrote:
Paul Heinlein wrote:
On Thu, 1 Dec 2005, rik@harddata.com wrote:
I am trying to install CentOS to a server that has a Tyan S2891 motherboard, 2 Opteron 875 Dual Cores and 2GB of RAM.
I've got a Tyan S2882-D with two dual-core Opteron 270s running CentOS 4 just fine -- but it's got the AMD-8111 chipset, not the nVidia CK804.
Since CentOS is based on stable/older code base, and Tyan S2891 is bleeding-edge-new, I wouldn't be surprised if it didn't work so well. My $0.02 ONLY, no 1st hand experience, YMMV, etc.
This is what I figured so I have tested with Fedora Core 4 installs.
Interesting to find that the 2.6.11 install kernel works but the latest 2.6.14 kernel has the same symptoms. I am scratching my head and firing up bugzilla.
I had similar problems with the Tyan B2891 'Transport GT24' - see:
https://www.redhat.com/archives/nahant-list/2005-November/msg00062.html
Also make sure that the 'Installed OS' is set to 'Linux' in the BIOS.
I am pretty sure that is set but I will check when I have access to the machine again.
Rik
I had similar problems with the Tyan B2891 'Transport GT24' - see:
https://www.redhat.com/archives/nahant-list/2005-November/msg00062.html
Also make sure that the 'Installed OS' is set to 'Linux' in the BIOS.
James Pearson
James,
Thanks for the tip. The 'Installed OS' was set to 'Other'. I didn't realize there were other settings as other Tyan boards have choices of 'Other' or 'DOS'. There were mystical settings for DOS, WIN95, LINUX, OTHER. 'LINUX' fixed my issue and it doesn't require boot parameters.
Rik
rik@harddata.com wrote:
Thanks for the tip. The 'Installed OS' was set to 'Other'. I didn't realize there were other settings as other Tyan
boards
have choices of 'Other' or 'DOS'. There were mystical
settings
for DOS, WIN95, LINUX, OTHER. 'LINUX' fixed my issue and
it
doesn't require boot parameters.
Welcome to Phoenix ServerBIOS. ;->
There's a reason why workstation and server mainboards use it, instead of the typical desktop mainboards with typical AMI BIOS.
Welcome to Phoenix ServerBIOS. ;->
There's a reason why workstation and server mainboards use it, instead of the typical desktop mainboards with typical AMI BIOS.
Any idea what the settings can actually change?
Cheers, MaZe.
Maciej ¯enczykowski maze@cela.pl wrote:
Any idea what the settings can actually change?
A crapload of things. ;->
A20 control, memory mappings above 1M, 16M, 64M, 1G, 4G, PCI bus control/mapping, ACPI features/defaults, etc..., etc..., etc...
As someone who regularly had issues loading OS/2 and NT on typical PC systems in the '90s, I quickly started to prefer Phoenix BIOS over AMI or Award.
Tyan's use of Phoenix ServerBIOS is why I prefer them, and they are very quick to deal with issues on new platforms -- of which Opteron quickly presented early on (regardless of OEM ;-).
"William A. Mahaffey III" wam@HiWAAY.net wrote:
Since CentOS is based on stable/older code base, and Tyan S2891 is bleeding-edge-new, I wouldn't be surprised if it didn't work so well. My $0.02 ONLY, no 1st hand experience, YMMV, etc.
The CK04 (nForce4/Pro/etc...) series is not much different than the earlier 02 (nForce2) and 03 (nForce3) series. So basically anything with kernel 2.4.23+ (including backports made to RHEL3 many updates ago) or 2.6.5+ will have _excellent_ support when it comes to I2C, HyperTransport-PCI interconnect, peripherals, etc... In fact, one could argue the best periperal support in Linux "out-of-the-box" these days is on the nForce2/3/4/Pro chipsets.
However, when you start looking at 2-3 on-board HyperTransport tunnels, then you're talking a _lot_ of bridging. E.g., the S2895 has not only the nForce Pro 2200 (with 4 PCIe masters, HT-native GbE and 4 SATA channels, plus the legacy PC), not only the nForce Pro 2050 (with 4 more PCIe masters, another HT-native GbE and 4 more SATA chennls), but also an AMD8131 (dual PCI-X 1.0 channels). That's a _lot_ of busses/peripherals to enumerate!
It's not the immaturity of the components -- it's the sheer number! They _all_ work just fine, because HyperTransport makes it possible. *BUT* automatic identification and enumeration of them all is the issue.
rik@harddata.com wrote:
I am trying to install CentOS to a server that has a Tyan S2891 motherboard, 2 Opteron 875 Dual Cores and 2GB of RAM. I have been able trying to install over a network but the tg3 driver will not load. I then installed from a DVD and
on
rebooting after the installation I cannot get the network drivers to load and the RAID controller in the PCI slot is
not
available. It seems that when I do an lspci -tv the only a minimal amount of devices are seen.
Yep, this has come up several times now (yet another thing to add to the ELManagers FAQ) -- not all PCI busses are getting enumerated and recognized for newer PCIe chipsets. Especially the up to 8 PCIe masters in nForce Pro 2200+2050, especially when there are also 2 more PCI-X channels c/o an AMD 8131 as well.
Remember, all AGP, PCI-X and PCIe busses/channels are, virtually, enumerated as PCI logic. Today's new generation of PCIe or PCIe+PCI-X have literally 6, 8 or even 10+ virtual PCI busses/channels compared to only 2-4 of the legacy AGP+PCI with optional PCI-X or two.
First off, get the absolute latest BIOS. Most issues do get resolved by it.
Secondly, try booting with a combination of ACPI/PCI kernel parameters. Using "pci=bios" might solve the problem, especially if you have the latest BIOS.
If you hit the archives regarding the Tyan S2895 (Thunder K8WE), you'll find the boot-time option that person used (I believe it was "pci=bios"?).
rik@harddata.com wrote:
I am trying to install CentOS to a server that has a Tyan S2891 motherboard, 2 Opteron 875 Dual Cores and 2GB of RAM. I have been able trying to install over a network but the tg3 driver will not load. I then installed from a DVD and
on
rebooting after the installation I cannot get the network drivers to load and the RAID controller in the PCI slot is
not
available. It seems that when I do an lspci -tv the only a minimal amount of devices are seen.
Yep, this has come up several times now (yet another thing to add to the ELManagers FAQ) -- not all PCI busses are getting enumerated and recognized for newer PCIe chipsets. Especially the up to 8 PCIe masters in nForce Pro 2200+2050, especially when there are also 2 more PCI-X channels c/o an AMD 8131 as well.
Remember, all AGP, PCI-X and PCIe busses/channels are, virtually, enumerated as PCI logic. Today's new generation of PCIe or PCIe+PCI-X have literally 6, 8 or even 10+ virtual PCI busses/channels compared to only 2-4 of the legacy AGP+PCI with optional PCI-X or two.
First off, get the absolute latest BIOS. Most issues do get resolved by it.
This is something I have done. I will check again for a newer one.
Secondly, try booting with a combination of ACPI/PCI kernel parameters. Using "pci=bios" might solve the problem, especially if you have the latest BIOS.
I was playing with ACPI boot options as in the past they have fixed various errors but I will try this too.
If you hit the archives regarding the Tyan S2895 (Thunder K8WE), you'll find the boot-time option that person used (I believe it was "pci=bios")
Thank you for the helpful post.
Rik
Secondly, try booting with a combination of ACPI/PCI kernel parameters. Using "pci=bios" might solve the problem, especially if you have the latest BIOS.
Thanks for the hints on using pci=. I got it to work. I couldn't find where pci=bios was documented but that option didn't work. I read in the kernel docs in regards to booting and tried pci=conf1 and this worked. The machine is up and running and doing well under testing. I don't know what conf1 means but I be darned if it doesn't work :)
If you hit the archives regarding the Tyan S2895 (Thunder K8WE), you'll find the boot-time option that person used (I believe it was "pci=bios"?).
Do you know where this option is documented?
Rik
rik@harddata.com wrote:
Thanks for the hints on using pci=. I got it to work. I couldn't find where pci=bios was documented but that option didn't work.
Boot Prompt HOWTO, Section 4: http://www.tldp.org/HOWTO/BootPrompt-HOWTO-4.html
I read in the kernel docs in regards to booting and tried pci=conf1 and this worked.
From the HOWTO ...
4.2 The `pci=bios' and `pci=nobios' Arguments These are used to set/clear the flag indicating that the PCI probing is to take place via the PCI BIOS. The default is to use the BIOS. 4.3 The `pci=conf1' and `pci=conf2' Arguments If PCI direct mode is enabled, the use of these enables either configuration Type 1 or Type 2. These implicitly clear the PCI BIOS probe flag (i.e. `pci=nobios') too.
So you actually want to use "pci=nobios" (my bad).
I think I remember talking about this in the previous S2895 thread, how the default changed from 2.4 to 2.6. In any case, I'm going to do more complete research before putting it in the ELManager's FAQ.
As I've told others, I don't always have the correct, exact answer, but I like to think I get people on the right track.
The machine is up and running and doing well under testing.
I don't know what conf1 means but I be darned if it doesn't work :)
It was the BIOS preventing Linux from enumerating the PCI busses and devices correctly. So by passing "pci=conf1", you effectively sent "pci=nobios" and blocked it. It was the exact opposite of what I was thinking.
Although it might be because 2.4 (CentOS 3) and 2.6 (CentOS 4) differ on defaults -- IIRC.
Bryan J. Smith wrote:
rik@harddata.com wrote:
Thanks for the hints on using pci=. I got it to work. I couldn't find where pci=bios was documented but that option didn't work.
Boot Prompt HOWTO, Section 4: http://www.tldp.org/HOWTO/BootPrompt-HOWTO-4.html
you could also look through the kernel-parameters.txt file in /usr/share/doc/kernel-doc-2.6.9/Documentation
it would have ( one hopes ) more relevant info w.r.t the presently running kernel.
- K
Bryan J. Smith wrote:
rik@harddata.com wrote:
Thanks for the hints on using pci=. I got it to work. I couldn't find where pci=bios was documented but that option didn't work.
Boot Prompt HOWTO, Section 4: http://www.tldp.org/HOWTO/BootPrompt-HOWTO-4.html
you could also look through the kernel-parameters.txt file in /usr/share/doc/kernel-doc-2.6.9/Documentation
I have looked through this directory rather quickly and have printed out the document you mentioned. I will have to make some time to read it today.
Rik
rik@harddata.com wrote:
someone-else> you could also look through the someone-else> kernel-parameters.txt file in someone-else> /usr/share/doc/kernel-doc-2.6.9/Documentation
I have looked through this directory rather quickly and have printed out the document you mentioned. I will have to make some time to read it today.
Yes, definitely hit the share doc's for the _latest_ info on your kernel. 2.4 and 2.6 have _wildly_differing_ defaults for _key_ boot-time parameters, let alone individual kernel revisions could have key changes/additions.
Excellent suggestion whoever made it.
rik@harddata.com wrote:
Thanks for the hints on using pci=. I got it to work. I couldn't find where pci=bios was documented but that option didn't work.
Boot Prompt HOWTO, Section 4: http://www.tldp.org/HOWTO/BootPrompt-HOWTO-4.html
I read in the kernel docs in regards to booting and tried pci=conf1 and this worked.
From the HOWTO ...
4.2 The `pci=bios' and `pci=nobios' Arguments These are used to set/clear the flag indicating that the PCI probing is to take place via the PCI BIOS. The default is to use the BIOS. 4.3 The `pci=conf1' and `pci=conf2' Arguments If PCI direct mode is enabled, the use of these enables either configuration Type 1 or Type 2. These implicitly clear the PCI BIOS probe flag (i.e. `pci=nobios') too.
This must be old documentation. The pci=nobios flag didn't work.
The problem was fixed by adjusting a BIOS setting. By changing the Installed OS to Linux I am able to boot the machine without any boot flags.
So you actually want to use "pci=nobios" (my bad).
Rik
rik@harddata.com wrote:
This must be old documentation. The pci=nobios flag didn't work.
It doesn't mean the documentation is old. The HOWTO clearly lists the "pci=conf1" and "pci=conf2" options. They wouldn't exist if they merely only sent "pci=nobios" -- they are doing something differently.
I probably means the "pci=conf1" does something a few things more than just what "pci=nobios" does. Most docs say try "pci=nobios", "pci=conf1" and "pci=conf2" to see if they resolve issues with how the BIOS enumerates PCI busses/devices.
It wouldn't surprise me if modern Phoenix ServerBIOSes have several ACPI configuration modes that send limited info by default, and offer 1-2 additional modes (hence conf1/conf2) with more info -- detailed lists that might hang OSes like Windows (especially DOS-based) which are rather simplistic and limited at the bootstrap.
Again, I'll do more research and put this in the ELManagers FAQ.
The problem was fixed by adjusting a BIOS setting. By changing the Installed OS to Linux I am able to boot the machine without any boot flags.
What this probably does is send the "Conf1" info as the default. Again, I need to research this more.
[ Although I'd love to see if a Windows install/boot hangs with that setting -- especially Windows 95/98. ;-]
rik@harddata.com wrote:
This must be old documentation. The pci=nobios flag didn't work.
It doesn't mean the documentation is old. The HOWTO clearly lists the "pci=conf1" and "pci=conf2" options. They wouldn't exist if they merely only sent "pci=nobios" -- they are doing something differently.
This makes sense.
I probably means the "pci=conf1" does something a few things more than just what "pci=nobios" does. Most docs say try "pci=nobios", "pci=conf1" and "pci=conf2" to see if they resolve issues with how the BIOS enumerates PCI busses/devices.
Now that I have success at making the machine work I need to get the project going so I have limited time to explore this.
It wouldn't surprise me if modern Phoenix ServerBIOSes have several ACPI configuration modes that send limited info by default, and offer 1-2 additional modes (hence conf1/conf2) with more info -- detailed lists that might hang OSes like Windows (especially DOS-based) which are rather simplistic and limited at the bootstrap.
Again, I'll do more research and put this in the ELManagers FAQ.
What this probably does is send the "Conf1" info as the default. Again, I need to research this more.
[ Although I'd love to see if a Windows install/boot hangs with that setting -- especially Windows 95/98. ;-]
I would like to experiment but I just don't have the time. Besides I think my Win98CD is underneath my crack I mean coffee mug right now. :)
Rik
-- Bryan J. Smith | Sent from Yahoo Mail mailto:b.j.smith@ieee.org | (please excuse any http://thebs413.blogspot.com/ | missing headers) _______________________________________________ CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
-
Bryan J. Smith -
James Pearson -
Karanbir Singh -
Maciej Żenczykowski -
Paul Heinlein -
rik＠harddata.com -
Robert -
William A. Mahaffey III