[-mm] Make the memory controller more desktop responsive

[-mm] Make the memory controller more desktop responsive

[Balbir Singh]

This patch makes the memory controller more responsive on my desktop.

Here is what the patch does

1. Reduces the number of retries to 2. We had 5 earlier, since we
   were controlling swap cache as well. We pushed data from mappings
   to swap cache and we needed additional passes to clear out the cache.
2. It sets all cached pages as inactive. We were by default marking
   all pages as active, thus forcing us to go through two passes for
   reclaiming pages
3. Removes congestion_wait(), since we already have that logic in
   do_try_to_free_pages()

Comments? Flames?

The patch is against 2.6.25-rc8-mm1

Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
---

 mm/memcontrol.c |    5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff -puN mm/memcontrol.c~memory-controller-improve-responsiveness mm/memcontrol.c
--- linux-2.6.25-rc8/mm/memcontrol.c~memory-controller-improve-responsiveness	2008-04-03 13:07:17.000000000 +0530
+++ linux-2.6.25-rc8-balbir/mm/memcontrol.c	2008-04-03 14:02:27.000000000 +0530
@@ -35,7 +35,7 @@
 #include <asm/uaccess.h>
 
 struct cgroup_subsys mem_cgroup_subsys;
-static const int MEM_CGROUP_RECLAIM_RETRIES = 5;
+static const int MEM_CGROUP_RECLAIM_RETRIES = 2;
 static struct kmem_cache *page_cgroup_cache;
 
 /*
@@ -604,7 +604,6 @@ retry:
 			mem_cgroup_out_of_memory(mem, gfp_mask);
 			goto out;
 		}
-		congestion_wait(WRITE, HZ/10);
 	}
 
 	pc->ref_cnt = 1;
@@ -612,7 +611,7 @@ retry:
 	pc->page = page;
 	pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
 	if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
-		pc->flags |= PAGE_CGROUP_FLAG_CACHE;
+		pc->flags = PAGE_CGROUP_FLAG_CACHE;
 
 	lock_page_cgroup(page);
 	if (page_get_page_cgroup(page)) {
_

-- 
	Warm Regards,
	Balbir Singh
	Linux Technology Center
	IBM, ISTL

Re: [-mm] Make the memory controller more desktop responsive

[KAMEZAWA Hiroyuki]

On Thu, 03 Apr 2008 15:02:53 +0530
Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> This patch makes the memory controller more responsive on my desktop.
> 
> Here is what the patch does
> 
> 1. Reduces the number of retries to 2. We had 5 earlier, since we
>    were controlling swap cache as well. We pushed data from mappings
>    to swap cache and we needed additional passes to clear out the cache.

Hmm, what this change improves ?
I don't want to see OOM.

> 2. It sets all cached pages as inactive. We were by default marking
>    all pages as active, thus forcing us to go through two passes for
>    reclaiming pages
Agreed.

> 3. Removes congestion_wait(), since we already have that logic in
>    do_try_to_free_pages()
> 
Agreed.

Thanks,
-Kame

Re: [-mm] Make the memory controller more desktop responsive

[Balbir Singh]

KAMEZAWA Hiroyuki wrote:
> On Thu, 03 Apr 2008 15:02:53 +0530
> Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
>> This patch makes the memory controller more responsive on my desktop.
>>
>> Here is what the patch does
>>
>> 1. Reduces the number of retries to 2. We had 5 earlier, since we
>>    were controlling swap cache as well. We pushed data from mappings
>>    to swap cache and we needed additional passes to clear out the cache.
> 
> Hmm, what this change improves ?
> I don't want to see OOM.
> 

I had set it to 5 earlier, since the swap cache came back to our memory
controller, where it was accounted. I have not seen OOM with it on my desktop,
but at some point if the memory required is so much that we cannot fulfill it,
we do OOM. I have not seen any OOM so far with these changes.

>> 2. It sets all cached pages as inactive. We were by default marking
>>    all pages as active, thus forcing us to go through two passes for
>>    reclaiming pages
> Agreed.
> 
>> 3. Removes congestion_wait(), since we already have that logic in
>>    do_try_to_free_pages()
>>
> Agreed.


-- 
	Warm Regards,
	Balbir Singh
	Linux Technology Center
	IBM, ISTL

Re: [-mm] Make the memory controller more desktop responsive

[KAMEZAWA Hiroyuki]

On Thu, 03 Apr 2008 15:14:32 +0530
Balbir Singh <balbir@linux.vnet.ibm.com> wrote:

> KAMEZAWA Hiroyuki wrote:
> > On Thu, 03 Apr 2008 15:02:53 +0530
> > Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> >> This patch makes the memory controller more responsive on my desktop.
> >>
> >> Here is what the patch does
> >>
> >> 1. Reduces the number of retries to 2. We had 5 earlier, since we
> >>    were controlling swap cache as well. We pushed data from mappings
> >>    to swap cache and we needed additional passes to clear out the cache.
> > 
> > Hmm, what this change improves ?
> > I don't want to see OOM.
> > 
> 
> I had set it to 5 earlier, since the swap cache came back to our memory
> controller, where it was accounted. I have not seen OOM with it on my desktop,
> but at some point if the memory required is so much that we cannot fulfill it,
> we do OOM. I have not seen any OOM so far with these changes.
> 
Hmm, I'm now testing swap-cache accounting patch.
It seems I should check this value again.

Thanks,
-Kame