Skip to contentSkip to conversation

Linux speed on sun4c

DISCUSSION Aaron Lehmann 4 participants 5 messages Jul 1, 2001 – Oct 21, 2004

lore
Share & tools
lore
#0 lore raw

NetBSD/Sparc's FAQ asserts:

Why is NetBSD so much faster than SparcLinux on sun4c (top)

The memory management hardware on sun4c machines (SPARCStation 1, 1+, 2, IPC, IPX, SLC, ELC and clones) is not handled particularly well by Linux. Until Linux reworks their MMU code NetBSD will be very much faster on this hardware.

Was there ever any truth to this statement? It seems to be light on technical details. Have these purported issues ever been fixed?

I don't want to be scared into running NetBSD on my SparcStation 2 :D.

#1 lore raw
Quoted 10 lines · Aaron Lehmann

NetBSD/Sparc's FAQ asserts:

Why is NetBSD so much faster than SparcLinux on sun4c (top)

The memory management hardware on sun4c machines (SPARCStation 1, 1+, 2, IPC, IPX, SLC, ELC and clones) is not handled particularly well by Linux. Until Linux reworks their MMU code NetBSD will be very much faster on this hardware.

Quoted 4 lines

Was there ever any truth to this statement? It seems to be light on technical details. Have these purported issues ever been fixed?

I don't want to be scared into running NetBSD on my SparcStation 2 :D.

It's totally true, use *BSD on your sun4c systems if top performance is your desire. :-)

I know how to fix it but frankly I have no desire to work on that platform any more.

Later,
David S. Miller
[email protected]

#2 lore raw
In reply to Aaron Lehmann
Quoted 10 lines · David S. Miller
Quoted 9 lines · Aaron Lehmann

NetBSD/Sparc's FAQ asserts:

Why is NetBSD so much faster than SparcLinux on sun4c (top)

The memory management hardware on sun4c machines (SPARCStation 1, 1+, 2, IPC, IPX, SLC, ELC and clones) is not handled particularly well by Linux. Until Linux reworks their MMU code NetBSD will be very much faster on this hardware.

Quoted 4 lines
Quoted 4 lines

Was there ever any truth to this statement? It seems to be light on technical details. Have these purported issues ever been fixed?

I don't want to be scared into running NetBSD on my SparcStation 2 :D.

What about OpenBSD? Same as NetBSD?

Quoted 9 lines

It's totally true, use *BSD on your sun4c systems if top performance is your desire. :-)

I know how to fix it but frankly I have no desire to work on that platform any more.

Later,
David S. Miller
[email protected]

David, can you publish your idea for a fix? Possibly anybody elese can make the patch?

Thanks, Guenter

Signature · 7 lines
Dipl.-Ing. Guenter Millahn         Brandenburg University of Technology
Systems, Network & DB Admin        CS Dept / DB & IS Research Group
Voice: +49 (355) 69-2272/2700      P.O. Box: 10 13 44
Fax:   +49 (355) 69-2766           D-03013 Cottbus              GERMANY

"The real world is still far away from be led ad absurdum by the virtual
one."    (Hal Faber, newsreel "What happened, what will be", 08/13/2000)
#3 lore raw
Quoted 2 lines · Guenter Millahn

David, can you publish your idea for a fix? Possibly anybody elese can make the patch?

Currently under Linux when a constext is recycled because a new context is needed but all are in use, we basically toss all of the MMU segments that context owned.

This is bogus because if the contexts are the limited resource not the MMU segments themselves, we take a lot of false MMU misses on each context switch for no reason.

The solution is to link the MMU segment software state structures into the mm_struct. When an 'mm' reacquires a hw context, if any MMU segments remain on the mm's list, just pluck them back into the MMU.

Later,
David S. Miller
[email protected]

#4 lore raw

Did anybody wrote a patch for that?

Nico

David S. Miller [Tue, Jul 03, 2001 at 04:07:16PM -0700]:

Quoted 27 lines
Quoted 2 lines · Guenter Millahn

David, can you publish your idea for a fix? Possibly anybody elese can make the patch?

Currently under Linux when a constext is recycled because a new context is needed but all are in use, we basically toss all of the MMU segments that context owned.

This is bogus because if the contexts are the limited resource not the MMU segments themselves, we take a lot of false MMU misses on each context switch for no reason.

The solution is to link the MMU segment software state structures into the mm_struct. When an 'mm' reacquires a hw context, if any MMU segments remain on the mm's list, just pluck them back into the MMU.

Later,
David S. Miller
[email protected]
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Signature · 3 lines
Keep it simple & stupid, use what's available.
Please use pgp encryption: 8D0E 27A4 is my id.
http://nico.schotteli.us | http://linux.schottelius.org