linux-mm.kvack.org archive mirror
 help / color / mirror / Atom feed
* RE: [RFC 2/3] LVHPT - Setup LVHPT
@ 2006-05-02 15:03 Luck, Tony
  2006-05-02 17:30 ` Chen, Kenneth W
  2006-05-02 21:29 ` Ian Wienand
  0 siblings, 2 replies; 8+ messages in thread
From: Luck, Tony @ 2006-05-02 15:03 UTC (permalink / raw)
  To: Ian Wienand, linux-ia64; +Cc: linux-mm

Ian,

Thanks for keeping this alive.  Previous measurements on long
format VHPT were mostly close to neutral performance-wise with
short format ... so this is still waiting for the killer-app in
the form of another patch that actually uses features of the
long format VHPT to do something that can't easily be done by
the short format to give me an incentive to complicate the code
by adding yet another CONFIG option.  In fact, I'd prefer to see
a compelling use case for long format so that it would be clear
that the right thing to do would be to just remove short format
and replace it with long format, but I don't expect that things
will ever be that simple :-(

+ 	help
+ 	  The long format VHPT is an alternative hashed page table. Advantages
+ 	  of the long format VHPT are lower memory usage when there are a large
+ 	  number of processes in the system.

Is this really true?  Don't you still have all of the 3-level (or 4-level)
tree allocated to keep the machine independent code in mm/memory.c
happy in addition to the big block of memory that you are using on
each cpu for the LVHPT?  Where is the saving?

-Tony

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

^ permalink raw reply	[flat|nested] 8+ messages in thread
* RE: [RFC 2/3] LVHPT - Setup LVHPT
@ 2006-05-04 16:58 Luck, Tony
  0 siblings, 0 replies; 8+ messages in thread
From: Luck, Tony @ 2006-05-04 16:58 UTC (permalink / raw)
  To: Ian Wienand, Chen, Kenneth W; +Cc: linux-ia64, linux-mm

> Being relatively inexperienced, all this dynamic patching (SMP, page
> table, this) scares me in that what is executing diverges from what
> appears to be in source code, making difficult things even more
> difficult to debug.  Is there consensus that a long term goal should
> be that short and long formats should be dynamically selectable?

I wouldn't rule anything out until I see what can be done, and how
maintainable the code to do it is.  Perhaps someone will come up with
the ultimate in dynamic selection and use long format for some processes,
and short format for others (and thus get around the objections that
some workloads perform less well with long format).

-Tony

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

^ permalink raw reply	[flat|nested] 8+ messages in thread
* RE: [RFC 2/3] LVHPT - Setup LVHPT
@ 2006-05-02 21:33 Luck, Tony
  0 siblings, 0 replies; 8+ messages in thread
From: Luck, Tony @ 2006-05-02 21:33 UTC (permalink / raw)
  To: Ian Wienand; +Cc: linux-ia64, linux-mm

> Yes that does seem a bit miss-leading.  I guess the point was that
> with short format you dedicate the top areas of your region to page
> tables for each process, with long format it is static.

So perhaps adding the word "virtual" (in between the "lower" and the
"memory") into the help description, and dropping the bit "when there
are a large number of processes in the system" would be clearer?

-Tony

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

^ permalink raw reply	[flat|nested] 8+ messages in thread
* [RFC 0/3] IA64 Long Format VHPT support
@ 2006-05-02  5:25 Ian Wienand
  2006-05-02  5:25 ` [RFC 2/3] LVHPT - Setup LVHPT Ian Wienand
  0 siblings, 1 reply; 8+ messages in thread
From: Ian Wienand @ 2006-05-02  5:25 UTC (permalink / raw)
  To: linux-ia64; +Cc: linux-mm, Ian Wienand

Hi,

Following from this message are patches to enable the Long Format VHPT
on IA64, which I am posting in the hope of community review.  They are
against 2.6.17-rc3, and work for machines I have access to.  These
patches have long been a chicken-egg problem, but I believe that there
are now multiple people interested in using LVHPT for dynamic page
size support in some form.

There are two papers which reference this work

Itanium Page Tables and TLB
Matthew Chapman, Ian Wienand, Gernot Heiser
http://citeseer.ist.psu.edu/chapman03itanium.html

Itanium - A System Implementor's Tale
Charles Gray, Matthew Chapman, Peter Chubb, David Mosberger-Tang, Gernot Heiser
http://www.usenix.org/events/usenix05/tech/general/gray/gray_html/index.html

Any comments are welcomed.

-i

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

^ permalink raw reply	[flat|nested] 8+ messages in thread

end of thread, other threads:[~2006-05-04 16:58 UTC | newest]

Thread overview: 8+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2006-05-02 15:03 [RFC 2/3] LVHPT - Setup LVHPT Luck, Tony
2006-05-02 17:30 ` Chen, Kenneth W
2006-05-03  7:49   ` Ian Wienand
2006-05-03  8:07     ` Christian Hildner
2006-05-02 21:29 ` Ian Wienand
  -- strict thread matches above, loose matches on Subject: below --
2006-05-04 16:58 Luck, Tony
2006-05-02 21:33 Luck, Tony
2006-05-02  5:25 [RFC 0/3] IA64 Long Format VHPT support Ian Wienand
2006-05-02  5:25 ` [RFC 2/3] LVHPT - Setup LVHPT Ian Wienand

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).