Hello.
I understand one of the biggest merits of LinuxBIOS is fast booting.
Then, how fast it is? The top page of the Wiki says 3 sec, and "7/25/01: Dual Athlon update " issue says about 2 sec to boot a kernel. But this log shows it took 20 sec. http://www.carbotpc.com/linux/boottiming.log.txt It is too big difference between 2 sec and 20 sec, dont you?
Especially, I would want to know the figures on VIA mini-ITX mothres with USB memory booting. It is very good that if I could get /bin/init started within 3 sec or such.
--- Okajima, Jun. Tokyo, Japan.
http://www.carbotpc.com/linux/boottiming.log.txt It is too big difference between 2 sec and 20 sec, dont you?
If you search the mailing list you will probally find a thread on that report. There is something wrong with his setup but rather than fix it they seem to just want to post thier results. 2 and 3 seconds is the norm.
Any one have any luck on fixing the EPIA-M code? -Adam
On Tue, 5 Apr 2005, Adam Talbot wrote:
Any one have any luck on fixing the EPIA-M code?
working on it, hang in there. It got pretty badly messed up.
ron
Thank you for the update. -Adam ----- Original Message ----- From: "Ronald G. Minnich" rminnich@lanl.gov To: "Adam Talbot" talbotx@comcast.net Cc: linuxbios@openbios.org Sent: Tuesday, April 12, 2005 8:46 AM Subject: Re: [LinuxBIOS] EPIA-M code still seems to be broken
On Tue, 5 Apr 2005, Adam Talbot wrote:
Any one have any luck on fixing the EPIA-M code?
working on it, hang in there. It got pretty badly messed up.
ron
Jun OKAJIMA wrote:
Hello.
I understand one of the biggest merits of LinuxBIOS is fast booting.
Then, how fast it is? The top page of the Wiki says 3 sec, and "7/25/01: Dual Athlon update " issue says about 2 sec to boot a kernel. But this log shows it took 20 sec.
The dominant factors are device driver init/probing and starting services. Thus, boot time will be very dependent on system configuration. For the the 3 second example I booted to a shell prompt using an initrd image, serial console and not much else. Most real systems are going to take a fair bit longer than that.
The main difference w/linuxbios is that the kernel is loaded and running in under 1 second vs. 10-50 seconds for a COTS bios. How fast things go once the kernel is booted is up to you.
Cheers! Ty
On Tue, 5 Apr 2005, Tyson Sawyer wrote:
Jun OKAJIMA wrote:
Hello.
I understand one of the biggest merits of LinuxBIOS is fast booting.
Then, how fast it is? The top page of the Wiki says 3 sec, and "7/25/01: Dual Athlon update " issue says about 2 sec to boot a kernel. But this log shows it took 20 sec.
The dominant factors are device driver init/probing and starting services. Thus, boot time will be very dependent on system configuration. For the the 3 second example I booted to a shell prompt using an initrd image, serial console and not much else. Most real systems are going to take a fair bit longer than that.
The main difference w/linuxbios is that the kernel is loaded and running in under 1 second vs. 10-50 seconds for a COTS bios. How fast things go once the kernel is booted is up to you.
Cheers! Ty
FAQ entry anyone?
ron
At what speed do I need to set my console output in linux bios to have it not effect boot time of linux bios? 19200, 9600, 4800? This is in standard debug, I dont need to see all the information on the screen. Just the key points. -Adam
115200 is really good. 9600 noticeably slows it down.
ron
On 4/13/05, Adam Talbot talbotx@comcast.net wrote:
At what speed do I need to set my console output in linux bios to have it not effect boot time of linux bios? 19200, 9600, 4800? This is in standard debug, I dont need to see all the information on the screen. Just the key points.
Well ouptut at any speed at all affects the boot time so as terse of output as you can stand is the best speed up. But if you want ot minimize the impact of the serial delay then set it as fast as possible. I run mine at 115200.
9600 baud = 1.042 mS per byte. 115200 baud = 86 uS per byte.
So @115200 in 1 second you can dump 11,500 characters vs 960 @ 9600.
Hummm. So if console slows boot time, how about loading the vga bios stuff. Does this slow down the overall boot time of linux bios? -Adam
----- Original Message ----- From: "Richard Smith" smithbone@gmail.com To: "Adam Talbot" talbotx@comcast.net Cc: "Ronald G. Minnich" rminnich@lanl.gov; linuxbios@openbios.org Sent: Wednesday, April 13, 2005 8:14 AM Subject: Re: [LinuxBIOS] Console debug (boot time)
On 4/13/05, Adam Talbot talbotx@comcast.net wrote:
At what speed do I need to set my console output in linux bios to have it not effect boot time of linux bios? 19200, 9600, 4800? This is in standard debug, I dont need to see all the information on the screen. Just the key points.
Well ouptut at any speed at all affects the boot time so as terse of output as you can stand is the best speed up. But if you want ot minimize the impact of the serial delay then set it as fast as possible. I run mine at 115200.
9600 baud = 1.042 mS per byte. 115200 baud = 86 uS per byte.
So @115200 in 1 second you can dump 11,500 characters vs 960 @ 9600.
On 4/13/05, Adam Talbot talbotx@comcast.net wrote:
Hummm. So if console slows boot time, how about loading the vga bios stuff. Does this slow down the overall boot time of linux bios? -Adam
Yep. Its got run the factory vgabios which has a lot of delays in it and writeing to video memory is slow in general although no where near as slow as writeing to the serial port.
Cool, all good things to know. I will post all this information, and all then numbers I can get under the FAQ, when we get a running version of the EPIA-M code. Thx -Adam ----- Original Message ----- From: "Richard Smith" smithbone@gmail.com To: "Adam Talbot" talbotx@comcast.net Cc: linuxbios@openbios.org Sent: Wednesday, April 13, 2005 10:56 AM Subject: Re: [LinuxBIOS] Console debug (boot time)
On 4/13/05, Adam Talbot talbotx@comcast.net wrote:
Hummm. So if console slows boot time, how about loading the vga bios stuff. Does this slow down the overall boot time of linux bios? -Adam
Yep. Its got run the factory vgabios which has a lot of delays in it and writeing to video memory is slow in general although no where near as slow as writeing to the serial port.
On Wed, 13 Apr 2005, Richard Smith wrote:
Yep. Its got run the factory vgabios which has a lot of delays in it and writeing to video memory is slow in general although no where near as slow as writeing to the serial port.
just watching it run, the video startup *seems* fast. Once video is running, it's way faster than serial.
ron
On Wed, 13 Apr 2005, Adam Talbot wrote:
Hummm. So if console slows boot time, how about loading the vga bios stuff. Does this slow down the overall boot time of linux bios?
not from what I have seen. It's fast.
ron
* Jun OKAJIMA okajima@digitalinfra.co.jp [050405 18:31]:
Then, how fast it is? The top page of the Wiki says 3 sec, and "7/25/01: Dual Athlon update " issue says about 2 sec to boot a kernel. But this log shows it took 20 sec. http://www.carbotpc.com/linux/boottiming.log.txt It is too big difference between 2 sec and 20 sec, dont you?
From the log you can see that Linux starts booting after 13.252secs.
This is the value the above 2-3 seconds have to be measured against.
Another thing that is needed to get to 2-3secs is to disable most debugging. With serial speed at 9600 you will spend an extra second for each kilobyte of debugging output. Go disable serial POST.
Disabling sound support in filo might be a win, I have not checked though.
4 seconds are spent in the Linux kernel for IDE probing. There are better mechanisms to probe for IDE devices than what Linux does. OpenBIOS and filo are two examples of how to do it a lot quicker. I think some of this went into kernel 2.6, or at least I saw patches floating around that pushed IDE probing below 1s.
Especially, I would want to know the figures on VIA mini-ITX mothres with USB memory booting. It is very good that if I could get /bin/init started within 3 sec or such.
Use an initrd and remove the IDE driver completely from Linux. That will speed things up a lot when using 2.4
Stefan
On Apr 5, 2005 2:47 PM, Stefan Reinauer stepan@openbios.org wrote:
Another thing that is needed to get to 2-3secs is to disable most debugging. With serial speed at 9600 you will spend an extra second for each kilobyte of debugging output. Go disable serial POST.
Why disable debugging output when you can just bump the serial speed up to 115200. Thats the way I've always run my setups. Then you still get debugging output but its plenty fast.
Although if you are going for raw speed then yeah disable the serial debugging output
On Wed, 6 Apr 2005, Jun OKAJIMA wrote:
I understand one of the biggest merits of LinuxBIOS is fast booting.
one of the least important. Actually the least important. It's nice but not the reason we wrote it.
Then, how fast it is? The top page of the Wiki says 3 sec, and "7/25/01: Dual Athlon update " issue says about 2 sec to boot a kernel. But this log shows it took 20 sec. http://www.carbotpc.com/linux/boottiming.log.txt It is too big difference between 2 sec and 20 sec, dont you?
seems like different hardware takes different times.
ron
-
Adam Talbot -
Jun OKAJIMA -
Richard Smith -
Ronald G. Minnich -
Stefan Reinauer -
Tyson Sawyer