Subject: linux-2.6: Kernel supports "only" 32 cpu's
Package: linux-2.6
Severity: wishlist


Hi,

The current Debian kernel supports 32 cpu's according to the
/boot/config file. The
RHEL/CentOS 5 kernel has CONFIG_NR_CPUS set to 255. Any chance the
Debian kernel
parameter can be adjusted to be a bit higher? I assume multi core
machines will
become common hardware during the Lenny lifecycle. 32-bit machines are
expected in

6 months.

Regards,

Rik

-- System Information:
Debian Release: lenny/sid
APT prefers testing
APT policy: (500, 'testing')
Architecture: amd64 (x86_64)

Kernel: Linux 2.6.26-trunk-amd64 (SMP w/2 CPU cores)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/bash


Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm




--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org

On Fri, Jul 18, 2008 at 03:17:49PM +0200, Rik Theys wrote:
> The current Debian kernel supports 32 cpu's according to the
> /boot/config file. The
> RHEL/CentOS 5 kernel has CONFIG_NR_CPUS set to 255. Any chance the
> Debian kernel
> parameter can be adjusted to be a bit higher? I assume multi core
> machines will
> become common hardware during the Lenny lifecycle. 32-bit machines are
> expected in
> 6 months.

Setting it to 255 needs 2MiB of image size and unswappable memory.

| text data bss dec hex filename
| 1345623 3148364 417112 4911099 4aeffb x86_64-255/vmlinux
| 1339887 380556 273880 1994323 1e6e53 x86_64-32/vmlinux

It adds 32 interrupt vectors (a 32 bytes or so) for each new cpu into
the data section, maybe this can be changed to dynamic allocation some
time as this does not look like a critical path.

I don't think this is exchange of different resources. I know there are
already machines with nearly 200 cores out there and even Xen choked on
them. Maybe setting it to 64 will be okay for now.

Bastian

--
It would be illogical to assume that all conditions remain stable.
-- Spock, "The Enterprise Incident", stardate 5027.3

Bastian Blank wrote:

Setting it to 255 needs 2MiB of image size and unswappable memory.

| text data bss dec hex filename
| 1345623 3148364 417112 4911099 4aeffb x86_64-255/vmlinux
| 1339887 380556 273880 1994323 1e6e53 x86_64-32/vmlinux

It adds 32 interrupt vectors (a 32 bytes or so) for each new cpu into
the data section, maybe this can be changed to dynamic allocation some
time as this does not look like a critical path.


I would think that for most of the x86_64 machines, a 2MB increase in
memory usage is acceptable? Most of them are probably equiped with at
least 1Gb.




I don't think this is exchange of different resources. I know there are
already machines with nearly 200 cores out there and even Xen choked on
them. Maybe setting it to 64 will be okay for now.


OK, setting it to 64 will certainly help during the Lenny lifecycle.

Thanks,

Rik

Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm




--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org

On Tue, Jul 22, 2008 at 07:30:03AM +0200, Rik Theys wrote:
> I would think that for most of the x86_64 machines, a 2MB increase in
> memory usage is acceptable? Most of them are probably equiped with at
> least 1Gb.

| # free -m
| total used free shared buffers cached
| Mem: 128 107 20 0 27 23
| -/+ buffers/cache: 56 71
| Swap: 255 0 255
| # uname -m
| x86_64

(Okay, it is a virtualised machine ...)

Bastian

--
"Get back to your stations!"
"We're beaming down to the planet, sir."
-- Kirk and Mr. Leslie, "This Side of Paradise",
stardate 3417.3



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org