We are down to questions that are definitely appropriate for the list.
"Jeffrey B. Layton" laytonjb@bellsouth.net writes:
Eric W Biederman wrote:
"Jeffrey B. Layton" laytonjb@bellsouth.net writes:
Could you list some pros and cons for a prospective customer that is thinking about switching to LinuxBIOS but they are used to the regular old BIOS?
The quick list.
Pros: Better serial console support. Better debuggability. Better manageability. (You can do everything under Linux) Quick boot times. Has been tested on big (1000 node) clusters. Vendors care about cluster issues. You get the source.
Cons: Boots differently. Harder to find people supporting it.
I've got a question about the serial console support and the better manageability. With LinuxBIOS can you put BIOS messages across the network to the master node from power on? Or do you have to wait until you get to Lilo (and then use something like netconsole in the kernel)?
You can do serial from power on. Over the network is a more interesting problem, and hasn't been fully solved yet. Currently I connect the serial port to a box with a lot of serial ports, and that box is network accessible.
NIC hardware is noticeably more complex than serial ports so getting a network console going from power on is still an open issue. I don't know that it has been tested much yet.
As for the manageability issue. How is the cluster more manageable with LinuxBIOS (I'm not being argumentative, but rather curious)?
For example, if I have a cluster that's pretty stable where I don't lose many nodes and I don't need to change the BIOS at all, then what does LinuxBIOS give me in terms of manageability?
Manageability is really about when things fail, and have problems, not when they work. But even when things are mostly working you can set your BIOS options the very first time from Linux, and trust they are getting set the same as every other node. You get a reliable network boot, with etherboot. You get a fast boot. You get messages over the serial console from power on.
Maybe some of these things are only really annoying when the don't exist and you have lots of hardware. For me I like the ability to do a full install including flashing the BIOS on 350 nodes in under 10 minutes.
But manageability comes down to the fact that the little things are getting fixed in LinuxBIOS so it is not a pain to work with, after the learning curve.
Eric
On Tue, 2002-09-17 at 07:38, Eric W Biederman wrote:
You can do serial from power on. Over the network is a more interesting problem, and hasn't been fully solved yet. Currently I connect the serial port to a box with a lot of serial ports, and that box is network accessible.
Eric, Do you know if there is an KVM switch equivalent for serial port ?? How do you monitor multiple LinuxBIOS boxes except using many minicom ??
Maybe some of these things are only really annoying when the don't exist and you have lots of hardware. For me I like the ability to do a full install including flashing the BIOS on 350 nodes in under 10 minutes.
How do you connect these 350 nodes "physically" to you console ?? Do you need many consoles or is there any think like "hub" the LAN ??
Ollie
ollie lho ollie@sis.com.tw writes:
On Tue, 2002-09-17 at 07:38, Eric W Biederman wrote:
You can do serial from power on. Over the network is a more interesting problem, and hasn't been fully solved yet. Currently I connect the serial port to a box with a lot of serial ports, and that box is network accessible.
Eric, Do you know if there is an KVM switch equivalent for serial port ??
Yes. Usually I believe they are called serial concentrators. We have a device here designed to be used with clusters called the icebox. It combines the serial concentrator function, and the power management functionality, into one device.
How do you monitor multiple LinuxBIOS boxes except using many minicom ??
I open multiple telnet connections to the icebox.
Maybe some of these things are only really annoying when the don't exist and you have lots of hardware. For me I like the ability to do a full install including flashing the BIOS on 350 nodes in under 10 minutes.
How do you connect these 350 nodes "physically" to you console ?? Do you need many consoles or is there any think like "hub" the LAN ??
A network accessible device that is accept serial cables like a hub on the LAN. And it plugs into the LAN.
It is really convinient when you are debugging those BIOS problems that only affect a small percent of your nodes. I can sit at my desk and work on machines that are miles away, or I can write a script that will cold boot the machine all night long to test booting.
Eric
On Mon, 16 Sep 2002, Eric W Biederman wrote:
ollie lho ollie@sis.com.tw writes:
How do you monitor multiple LinuxBIOS boxes except using many minicom ??
I open multiple telnet connections to the icebox.
Here is a utility that manages interactive console sessions and logging via that type of device: http://www.llnl.gov/linux/conman/
Jim
On 17 Sep 2002, ollie lho wrote:
Do you know if there is an KVM switch equivalent for serial port ?? How do you monitor multiple LinuxBIOS boxes except using many minicom ??
The heck with that! Use my netconsole device. I have been using this for months and it is really wonderful. I can even catch kernel "oops" messages.
ron
Ronald G Minnich wrote:
On 17 Sep 2002, ollie lho wrote:
Do you know if there is an KVM switch equivalent for serial port ?? How do you monitor multiple LinuxBIOS boxes except using many minicom ??
Ron,
Could you post the netconsole patch for a recent kernel somewhere so I can try it?
Thanks!
Jeff
The heck with that! Use my netconsole device. I have been using this for months and it is really wonderful. I can even catch kernel "oops" messages.
ron
Linuxbios mailing list Linuxbios@clustermatic.org http://www.clustermatic.org/mailman/listinfo/linuxbios
There are two versions. The "known good" version is a module. You insmod it with parameters just like the original. It doesn't actually start logging over IP until you open the device -- this is for systems where the module might get loaded before IP is ready.
The second version has one additional wrinkle: it sends out broadcast MAC packets until IP is ready (i.e. until you open it) then sends out over IP.
The second version is not well tested. But you can have that one if you wish.
Let me know.
ron
oops, missed this message and sent a reply to a reply!
On 16 Sep 2002, Eric W Biederman wrote:
NIC hardware is noticeably more complex than serial ports so getting a network console going from power on is still an open issue. I don't know that it has been tested much yet.
we're going to try this using myrinet hardware. I.e. a very simple control program on the NIC that will give us "serial port" in linuxbios for the very early startup.
ron
Ronald G Minnich rminnich@lanl.gov writes:
oops, missed this message and sent a reply to a reply!
On 16 Sep 2002, Eric W Biederman wrote:
NIC hardware is noticeably more complex than serial ports so getting a network console going from power on is still an open issue. I don't know that it has been tested much yet.
we're going to try this using myrinet hardware. I.e. a very simple control program on the NIC that will give us "serial port" in linuxbios for the very early startup.
Sounds good. And I guess with hardware like myrinet with it's own processor and ram you can do a lot of interesting things.
Right now the serial port hardware has the advantage on PC's in that it is on the same bus as the rom chip. The one part of the machine that is alive and working at power on.
I could not debug my P64H2 pci lockup fix with anything on the pci bus. And I cannot use any NIC that uses DMA until the Ram is initialized. The bottleneck is both DMA'ing from ROM which might be possible, and getting the MAC address (or equivalent) into some place you can DMA from.
If I had to work with a simple chipset with very little redundancy then I guess a serial port would not have a real advantage. But while my best option are Intels 12 dozen hub monsters I like having a console on the most reliable path.
This is a BIOS developer perspective. But it is so cool to have some one else tell you where your BIOS locks up do to hardware bugs. Today there is a mature serial port infrastructure that works well, and I really like it.
I suspect I will like network consoles if I ever get beyond debugging BIOS issues. And I certainly support network consoles.
Has anyone else on the list had to fix problems where components with the same part number do not behave identically, but the both components were behaving correctly?
Eric
On 16 Sep 2002, Eric W Biederman wrote:
Has anyone else on the list had to fix problems where components with the same part number do not behave identically, but the both components were behaving correctly?
we have a bunch of compaq SP 750 nodes. Same part # right down to the chips. But depending on whether the chips are korean- or phillipine-made, you get a 10% difference in PCI bandwidth from Myrinet.
ron
Hi,
I am in the process of porting the Linux BIOS code for my system.
The Specification of my system is as follow: Main board: Intel vc15. SuperIO : SMSC LPC47M192 Northbridge: Intel 845 MCH (DDR support) Southbridge: Intel 82801BA (ICH2)
I have added source code to support the above mentioned component as per the specification of corresponding component. I have built the Linux BIOS image also. I need to test the code.
Before that I want to know, whether I need to apply any patches to the Linux kernel. I am using 2.4.18 version of Linux kernel. I am planning to boot the kernel from IDE disk.
"Please ignore the attachment!!!"
Thanks, Siva.s
-----Original Message----- From: linuxbios-admin@clustermatic.org [mailto:linuxbios-admin@clustermatic.org] On Behalf Of Ronald G Minnich Sent: Tuesday, September 17, 2002 7:28 AM To: Eric W Biederman Cc: Jeffrey B. Layton; LinuxBIOS Subject: Re: General LinuxBIOS Questions
oops, missed this message and sent a reply to a reply!
On 16 Sep 2002, Eric W Biederman wrote:
NIC hardware is noticeably more complex than serial ports so getting a network console going from power on is still an open issue. I don't know that it has been tested much yet.
we're going to try this using myrinet hardware. I.e. a very simple control program on the NIC that will give us "serial port" in linuxbios for the very early startup.
ron
_______________________________________________ Linuxbios mailing list Linuxbios@clustermatic.org http://www.clustermatic.org/mailman/listinfo/linuxbios
On Fri, 20 Sep 2002, SIVAKUMAR SUBRAMANI wrote:
Before that I want to know, whether I need to apply any patches to the Linux kernel. I am using 2.4.18 version of Linux kernel. I am planning to boot the kernel from IDE disk.
I always test with unpatched kernels the first time, to see if there are unresolved linuxbios issues.
ron
-
Eric W Biederman -
Jeffrey B. Layton -
Jim Garlick -
ollie lho -
Ronald G Minnich -
SIVAKUMAR SUBRAMANI