* [LSF/MM TOPIC] In-kernel compression in the MM subsystem
@ 2013-01-26 20:16 Dan Magenheimer
2013-02-01 1:46 ` Simon Jeons
2013-02-06 2:18 ` Simon Jeons
0 siblings, 2 replies; 6+ messages in thread
From: Dan Magenheimer @ 2013-01-26 20:16 UTC (permalink / raw)
To: lsf-pc; +Cc: linux-mm, Seth Jennings, Nitin Gupta, Konrad Wilk, Minchan Kim
There's lots of interesting things going on in kernel memory
management, but one only(?) increases the effective amount
of data that can be stored in a fixed amount of RAM: in-kernel
compression.
Since ramzswap/compcache (now zram) was first proposed in 2009
as an in-memory compressed swap device, there have been a number
of in-kernel compression solutions proposed, including
zcache, kztmem, and now zswap. Each shows promise to improve
performance by using compression under memory pressure to
reduce I/O due to swapping and/or paging. Each is still
in staging (though zram may be promoted by LSFMM 2013)
because each also brings a number of perplexing challenges.
I think it's time to start converging on which one or more
of these solutions, if any, should be properly promoted and
more fully integrated into the kernel memory management
subsystem. Before this can occur, it's important to build a
broader understanding and, hopefully, also a broader consensus
among the MM community on a number of key challenges and questions
in order to guide and drive further development and merging.
I would like to collect a list of issues/questions, and
start a discussion at LSF/MM by presenting this list, select
the most important, then lead a discussion on how ever many
there is time for. Most likely this is an MM-only discussion
though a subset might be suitable for a cross-talk presentataion.
Thanks!
Dan Magenheimer
LSF/MM attendee 2010,2011,2012
LSF/MM presenter (MM track) 2011,2012
--
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] 6+ messages in thread
* Re: [LSF/MM TOPIC] In-kernel compression in the MM subsystem
2013-01-26 20:16 [LSF/MM TOPIC] In-kernel compression in the MM subsystem Dan Magenheimer
@ 2013-02-01 1:46 ` Simon Jeons
2013-02-06 2:18 ` Simon Jeons
1 sibling, 0 replies; 6+ messages in thread
From: Simon Jeons @ 2013-02-01 1:46 UTC (permalink / raw)
To: Dan Magenheimer
Cc: lsf-pc, linux-mm, Seth Jennings, Nitin Gupta, Konrad Wilk,
Minchan Kim
Hi Dan,
On Sat, 2013-01-26 at 12:16 -0800, Dan Magenheimer wrote:
> There's lots of interesting things going on in kernel memory
> management, but one only(?) increases the effective amount
> of data that can be stored in a fixed amount of RAM: in-kernel
> compression.
>
> Since ramzswap/compcache (now zram) was first proposed in 2009
> as an in-memory compressed swap device, there have been a number
> of in-kernel compression solutions proposed, including
> zcache, kztmem, and now zswap. Each shows promise to improve
What's the difference between in-memory compression and in-kernel
compression?
> performance by using compression under memory pressure to
> reduce I/O due to swapping and/or paging. Each is still
> in staging (though zram may be promoted by LSFMM 2013)
> because each also brings a number of perplexing challenges.
>
> I think it's time to start converging on which one or more
> of these solutions, if any, should be properly promoted and
> more fully integrated into the kernel memory management
> subsystem. Before this can occur, it's important to build a
> broader understanding and, hopefully, also a broader consensus
> among the MM community on a number of key challenges and questions
> in order to guide and drive further development and merging.
>
> I would like to collect a list of issues/questions, and
> start a discussion at LSF/MM by presenting this list, select
> the most important, then lead a discussion on how ever many
> there is time for. Most likely this is an MM-only discussion
> though a subset might be suitable for a cross-talk presentataion.
>
> Thanks!
> Dan Magenheimer
> LSF/MM attendee 2010,2011,2012
> LSF/MM presenter (MM track) 2011,2012
>
>
> --
> 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>
--
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] 6+ messages in thread
* Re: [LSF/MM TOPIC] In-kernel compression in the MM subsystem
2013-01-26 20:16 [LSF/MM TOPIC] In-kernel compression in the MM subsystem Dan Magenheimer
2013-02-01 1:46 ` Simon Jeons
@ 2013-02-06 2:18 ` Simon Jeons
2013-02-06 18:40 ` Dan Magenheimer
1 sibling, 1 reply; 6+ messages in thread
From: Simon Jeons @ 2013-02-06 2:18 UTC (permalink / raw)
To: Dan Magenheimer
Cc: lsf-pc, linux-mm, Seth Jennings, Nitin Gupta, Konrad Wilk,
Minchan Kim
Hi Dan,
On Sat, 2013-01-26 at 12:16 -0800, Dan Magenheimer wrote:
> There's lots of interesting things going on in kernel memory
> management, but one only(?) increases the effective amount
> of data that can be stored in a fixed amount of RAM: in-kernel
> compression.
>
> Since ramzswap/compcache (now zram) was first proposed in 2009
> as an in-memory compressed swap device, there have been a number
> of in-kernel compression solutions proposed, including
> zcache, kztmem, and now zswap. Each shows promise to improve
> performance by using compression under memory pressure to
> reduce I/O due to swapping and/or paging. Each is still
> in staging (though zram may be promoted by LSFMM 2013)
> because each also brings a number of perplexing challenges.
>
> I think it's time to start converging on which one or more
> of these solutions, if any, should be properly promoted and
> more fully integrated into the kernel memory management
> subsystem. Before this can occur, it's important to build a
> broader understanding and, hopefully, also a broader consensus
> among the MM community on a number of key challenges and questions
> in order to guide and drive further development and merging.
>
> I would like to collect a list of issues/questions, and
> start a discussion at LSF/MM by presenting this list, select
> the most important, then lead a discussion on how ever many
> there is time for. Most likely this is an MM-only discussion
> though a subset might be suitable for a cross-talk presentataion.
>
Is there benchmark to test each component in tmem?
> Thanks!
> Dan Magenheimer
> LSF/MM attendee 2010,2011,2012
> LSF/MM presenter (MM track) 2011,2012
>
>
> --
> 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>
--
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] 6+ messages in thread
* RE: [LSF/MM TOPIC] In-kernel compression in the MM subsystem
2013-02-06 2:18 ` Simon Jeons
@ 2013-02-06 18:40 ` Dan Magenheimer
2013-02-13 8:08 ` Simon Jeons
0 siblings, 1 reply; 6+ messages in thread
From: Dan Magenheimer @ 2013-02-06 18:40 UTC (permalink / raw)
To: Simon Jeons
Cc: lsf-pc, linux-mm, Seth Jennings, Nitin Gupta, Konrad Wilk,
Minchan Kim
> From: Simon Jeons [mailto:simon.jeons@gmail.com]
> Subject: Re: [LSF/MM TOPIC] In-kernel compression in the MM subsystem
>
> Hi Dan,
> On Sat, 2013-01-26 at 12:16 -0800, Dan Magenheimer wrote:
> > There's lots of interesting things going on in kernel memory
> > management, but one only(?) increases the effective amount
> > of data that can be stored in a fixed amount of RAM: in-kernel
> > compression.
> >
> > Since ramzswap/compcache (now zram) was first proposed in 2009
> > as an in-memory compressed swap device, there have been a number
> > of in-kernel compression solutions proposed, including
> > zcache, kztmem, and now zswap. Each shows promise to improve
> > performance by using compression under memory pressure to
> > reduce I/O due to swapping and/or paging. Each is still
> > in staging (though zram may be promoted by LSFMM 2013)
> > because each also brings a number of perplexing challenges.
> >
> > I think it's time to start converging on which one or more
> > of these solutions, if any, should be properly promoted and
> > more fully integrated into the kernel memory management
> > subsystem. Before this can occur, it's important to build a
> > broader understanding and, hopefully, also a broader consensus
> > among the MM community on a number of key challenges and questions
> > in order to guide and drive further development and merging.
> >
> > I would like to collect a list of issues/questions, and
> > start a discussion at LSF/MM by presenting this list, select
> > the most important, then lead a discussion on how ever many
> > there is time for. Most likely this is an MM-only discussion
> > though a subset might be suitable for a cross-talk presentataion.
> >
>
> Is there benchmark to test each component in tmem?
Hi Simon --
I'm not sure what you mean. Could you add a few words
to clarify?
Thanks,
Dan
--
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] 6+ messages in thread
* Re: [LSF/MM TOPIC] In-kernel compression in the MM subsystem
2013-02-06 18:40 ` Dan Magenheimer
@ 2013-02-13 8:08 ` Simon Jeons
2013-02-13 16:47 ` Dan Magenheimer
0 siblings, 1 reply; 6+ messages in thread
From: Simon Jeons @ 2013-02-13 8:08 UTC (permalink / raw)
To: Dan Magenheimer
Cc: lsf-pc, linux-mm, Seth Jennings, Nitin Gupta, Konrad Wilk,
Minchan Kim
On Wed, 2013-02-06 at 10:40 -0800, Dan Magenheimer wrote:
> > From: Simon Jeons [mailto:simon.jeons@gmail.com]
> > Subject: Re: [LSF/MM TOPIC] In-kernel compression in the MM subsystem
> >
> > Hi Dan,
> > On Sat, 2013-01-26 at 12:16 -0800, Dan Magenheimer wrote:
> > > There's lots of interesting things going on in kernel memory
> > > management, but one only(?) increases the effective amount
> > > of data that can be stored in a fixed amount of RAM: in-kernel
> > > compression.
> > >
> > > Since ramzswap/compcache (now zram) was first proposed in 2009
> > > as an in-memory compressed swap device, there have been a number
> > > of in-kernel compression solutions proposed, including
> > > zcache, kztmem, and now zswap. Each shows promise to improve
> > > performance by using compression under memory pressure to
> > > reduce I/O due to swapping and/or paging. Each is still
> > > in staging (though zram may be promoted by LSFMM 2013)
> > > because each also brings a number of perplexing challenges.
> > >
> > > I think it's time to start converging on which one or more
> > > of these solutions, if any, should be properly promoted and
> > > more fully integrated into the kernel memory management
> > > subsystem. Before this can occur, it's important to build a
> > > broader understanding and, hopefully, also a broader consensus
> > > among the MM community on a number of key challenges and questions
> > > in order to guide and drive further development and merging.
> > >
> > > I would like to collect a list of issues/questions, and
> > > start a discussion at LSF/MM by presenting this list, select
> > > the most important, then lead a discussion on how ever many
> > > there is time for. Most likely this is an MM-only discussion
> > > though a subset might be suitable for a cross-talk presentataion.
> > >
> >
> > Is there benchmark to test each component in tmem?
>
> Hi Simon --
>
> I'm not sure what you mean. Could you add a few words
> to clarify?
>
Hi Dan,
Some questions about zsmalloc:
1) What's the meaning of comment above USE_PGTABLE_MAPPING macro "This
cause zsmalloc to use page table mapping rather than copying for object
mapping"?
2) How zsmalloc handle object span two pages? It seems that in function
init_zspage, link->next = obj_location_to_handle(next_page, 0); you
encode next_page and 0 to object, then how can zs_malloc find this free
object? IIUC, this encode skip the object span two pages.
3) Why must map after malloc if want to use a object?
4) What's the number of ZS_MAX_ALLOC_SIZE and ZS_MIN_ALLOC_SIZE? There
are too many macros to figure it out.
Thanks,
Simon
> Thanks,
> Dan
--
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] 6+ messages in thread
* RE: [LSF/MM TOPIC] In-kernel compression in the MM subsystem
2013-02-13 8:08 ` Simon Jeons
@ 2013-02-13 16:47 ` Dan Magenheimer
0 siblings, 0 replies; 6+ messages in thread
From: Dan Magenheimer @ 2013-02-13 16:47 UTC (permalink / raw)
To: Simon Jeons
Cc: lsf-pc, linux-mm, Seth Jennings, Nitin Gupta, Konrad Wilk,
Minchan Kim
> From: Simon Jeons [mailto:simon.jeons@gmail.com]
> Subject: Re: [LSF/MM TOPIC] In-kernel compression in the MM subsystem
>
> On Wed, 2013-02-06 at 10:40 -0800, Dan Magenheimer wrote:
> > > From: Simon Jeons [mailto:simon.jeons@gmail.com]
> > > Subject: Re: [LSF/MM TOPIC] In-kernel compression in the MM subsystem
> > >
> > > Hi Dan,
> > > On Sat, 2013-01-26 at 12:16 -0800, Dan Magenheimer wrote:
> > > > There's lots of interesting things going on in kernel memory
> > > > management, but one only(?) increases the effective amount
> > > > of data that can be stored in a fixed amount of RAM: in-kernel
> > > > compression.
> > > >
> > > > Since ramzswap/compcache (now zram) was first proposed in 2009
> > > > as an in-memory compressed swap device, there have been a number
> > > > of in-kernel compression solutions proposed, including
> > > > zcache, kztmem, and now zswap. Each shows promise to improve
> > > > performance by using compression under memory pressure to
> > > > reduce I/O due to swapping and/or paging. Each is still
> > > > in staging (though zram may be promoted by LSFMM 2013)
> > > > because each also brings a number of perplexing challenges.
> > > >
> > > > I think it's time to start converging on which one or more
> > > > of these solutions, if any, should be properly promoted and
> > > > more fully integrated into the kernel memory management
> > > > subsystem. Before this can occur, it's important to build a
> > > > broader understanding and, hopefully, also a broader consensus
> > > > among the MM community on a number of key challenges and questions
> > > > in order to guide and drive further development and merging.
> > > >
> > > > I would like to collect a list of issues/questions, and
> > > > start a discussion at LSF/MM by presenting this list, select
> > > > the most important, then lead a discussion on how ever many
> > > > there is time for. Most likely this is an MM-only discussion
> > > > though a subset might be suitable for a cross-talk presentataion.
> > > >
> > >
> > > Is there benchmark to test each component in tmem?
> >
> > Hi Simon --
> >
> > I'm not sure what you mean. Could you add a few words
> > to clarify?
> >
>
> Hi Dan,
>
> Some questions about zsmalloc:
>
> 1) What's the meaning of comment above USE_PGTABLE_MAPPING macro "This
> cause zsmalloc to use page table mapping rather than copying for object
> mapping"?
> 2) How zsmalloc handle object span two pages? It seems that in function
> init_zspage, link->next = obj_location_to_handle(next_page, 0); you
> encode next_page and 0 to object, then how can zs_malloc find this free
> object? IIUC, this encode skip the object span two pages.
> 3) Why must map after malloc if want to use a object?
> 4) What's the number of ZS_MAX_ALLOC_SIZE and ZS_MIN_ALLOC_SIZE? There
> are too many macros to figure it out.
Hi Simon --
Those are good questions, but you are asking the wrong person.
I stopped using zsmalloc in zcache (and ramster) because it
has certain issues and it didn't appear to me those issues
would/could be resolved. I wrote a custom allocator (see zbud.c)
to avoid those issues. This is explained in:
http://lkml.indiana.edu/hypermail/linux/kernel/1208.1/03763.html
I'd be interested in hearing your ideas/requirements for
in-kernel compression!
Thanks,
Dan
P.S. Note that "old" zcache (which uses zsmalloc) will be gone in
3.8 and "new" zcache (which uses zbud and merges in ramster)
will be in drivers/staging/zcache in 3.8.
--
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] 6+ messages in thread
end of thread, other threads:[~2013-02-13 16:49 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2013-01-26 20:16 [LSF/MM TOPIC] In-kernel compression in the MM subsystem Dan Magenheimer
2013-02-01 1:46 ` Simon Jeons
2013-02-06 2:18 ` Simon Jeons
2013-02-06 18:40 ` Dan Magenheimer
2013-02-13 8:08 ` Simon Jeons
2013-02-13 16:47 ` Dan Magenheimer
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).