[Ocfs2-users] High on buffers and deep on swap

Eckenfels. Bernd B.Eckenfels at seeburger.de
Tue Apr 10 11:11:34 PDT 2007


Hello Luis,
 
the large usage of cache by inodes is not an issue in the typical
database setup, because you have only a very limited number of files
(used) on your filesystems. Also the database usually grows initially
and then does not request much more additional memory in operations.
 
However you are right, Oracle with Data Files on ocfsv2 has no need to
cache the data blocks on the FS layer. But I gues sthis is not done
(memadvice, fcntl or similiar methods can be used to enforce "directio"
- i am not sure which one is used by Oracle). I think the original
problem reports with large usage have been due to non-db workloads.
 
Gruss
Bernd

________________________________

From: ocfs2-users-bounces at oss.oracle.com
[mailto:ocfs2-users-bounces at oss.oracle.com] On Behalf Of Luis Freitas
Sent: Tuesday, April 10, 2007 7:59 PM
To: Alexei_Roudnev; ocfs2-users at oss.oracle.com
Subject: Re: [Ocfs2-users] High on buffers and deep on swap


Alexei,
 
    I am not a Linux kernel specialist, but generally memory on the
"buffer cache" is on a different "list" than memory "free", so the
kernel has to reclaim the buffers from the buffer "list" to release them
to the "free" "list" as memory pressure increases.
 
   This has a processing cost. And on large database servers is
undesirable as the database does not use the buffer cache, so we have
the processing cost without any benefit.
 
     Usually on a database environment I tune the kernel as to keep a
large amount of the memory on the "free" list, so that it be readly
available when spawning new database processes.
 
Regards,
Luis

Alexei_Roudnev <Alexei_Roudnev at exigengroup.com> wrote:

	Suni, you do know it much better - is not is strange a little?
(so many ext3 inode chache objects, togetrher with 3 GB of cached disk
space).
	 
	But I dont see anything wrong below - big 'cached' memory means
only 'you have so many unused memory that system cached files in it' and
nothing more. Cached memory is in reality _free for immediate use_
memory (can be used by anyone immediately).
	 
	What ysstem is doing is _instead of freeing memory and filling
it by zerous, free unused memory and fill it in by buffer cache data, so
that if someone need these data he have them at once. If someone need
more memory, system reuse any of 'chached' memory without any delay
(because these are buffers which are already written to the disk or
which was never udpated).
	 
	Big 'cached' value means only _big free memory_.
	 

		----- O  riginal Message ----- 
		From: Luis Freitas <mailto:lfreitas34 at yahoo.com>  
		To: Alexei_Roudnev
<mailto:Alexei_Roudnev at exigengroup.com>  ; Brian Sieler
<mailto:blsieler at gmail.com>  ; ocfs2-users at oss.oracle.com 
		Sent: Monday, April 09, 2007 4:32 PM
		Subject: Re: [Ocfs2-users] High on buffers and deep on
swap

		Alexei,
		 
		   How can I relate the information on slabtop to the
actual memory used by buffers?
		 
		   I see this on slabtop:
		 
		 Active / Total Objects (% used)    : 603822 / 649643
(92.9%)
		 Active / Total Slabs (% used)      : 47216 / 47216
(100.0%)
		 Active / Total Caches (% used)     : 97 / 133 (72.9%)
		 Active / Total Size (% used)       : 176601.08K /
181508.49K (97.3%)
		 Minimum / Average / Maximum Object : 0.01K / 0.28K /
128.00K
		  OBJS ACTIVE  USE OBJ SIZE  SLABS OBJ/SLAB CACHE SIZE
NAME
		202461 202451  99%    0.54K  28923        7    115692K
ext3_inode_cache
		232206 232153  99%    0.15K   8931       26     35724K
dentry_cache
		 29974  26092  87%    0.27K   2141       14      8564K
radix_tree_node
		 68250  62400  91%    0.05K    910       75      3640K
buffer_head
		   855    855 100%    4.00K    855        1      3420K
pmd
		   647    647 100%    4.00K    647        1      2588K
size-4096
		  8595   7694  89%    0.25K    573       15      2292K
filp
		 20835  17331  83%    0.09K    463       45      1852K
vm_area_struct
		   780    767  98%    2.00K    390        2      1560K
size-2048
		 18849   7962  42%    0.06K    309       61      1236K
size-64
		  2440    632  25%    0.50K    305        8      1220K
size-512
		  2926   2889  98%    0.34K    266       11      1064K
inode_cache
		   256    256 100%    3.00K    128        2      1024K
biovec-(256)
		   600    592  98%    1.38K    120        5       960K
task_struct
		   515    512  99%    1.38K    103        5       824K
pirpIo
		  5084   2365  46%    0.12K    164       31       656K
size-128
		
		 
		  The largest area is about 100Mb, but on free there are
over 3Gb on the "cached" column:
		 
		[oracle at br001sv0431 ~]$ free
		             total       used       free     shared
buffers     cached
		Mem:       5190736    4461420     729316          0
141836    3265464
		-/+ buffers/cache:    1054120    4136616
		Swap:      2048248          0    2048248
		[oracle at br001sv0431 ~]$
		
		 
		Regards,
		Luis
		 


		Alexei_Roudnev <Alexei_Roudnev at exigengroup.com> wrote:

			Did you run slabtop ? It can show unreleased
buffers in the system.
			 

				----- Original Message ----- 
				From: Luis Freitas
<mailto:lfreitas34 at yahoo.com>  
				To: Alexei_Roudnev
<mailto:Alexei_Roudnev at exigengroup.com>  ; Brian Sieler
<mailto:blsieler at gmail.com>  ; ocfs2-users at oss.oracle.com 
				Sent: Monday, April 09, 2007 2:29 PM
				Subject: Re: [Ocfs2-users] High on
buffers and deep on swap

				Alexei,
				 
				  Yes, it seems to have no effect, which
too is very strange. On 2.4 vm.freepages had a very easy to notice
effect.
				 
				   There are other people having
problems with buffers not being released on the list and some of them
are forcing the kernel cache to be flushed with:
				 
				echo 3 > /proc/sys/vm/drop_caches
				 
				    But I dont see this parameter on
RHAS 4.0.
				 
				     Also, to be fair this seems to be a
generic VM issue, I see this on servers that are not running ocfs2 too.
And I only see this behavior on machines with more than 2Gb of memory.
				 
				Regards,
				Luis
				 
				Regards,
				Luis
				
				Alexei_Roudnev
<Alexei_Roudnev at exigengroup.com> wrote:

				Did you tried vm.swappiness parameter?
				 
				(/proc/sys/vm/swappiness)
				 

				----- 
				Original Message ----- 
				From: Brian Sieler
<mailto:blsieler at gmail.com>  
				To: 'Luis Freitas'
<mailto:lfreitas34 at yahoo.com>  ; ocfs2-users at oss.oracle.com 
				Sent: Sunday, April 08, 2007 10:52 PM
				Subject: RE: [Ocfs2-users] High on
buffers and deep on swap

				Luis, yes I am experiencing what appears
to be a similar problem you are describing. See my post from just a few
minutes ago on another thread.
				 
				I run a 2-node cluster with OCFS2/RAC on
2.6.9-34.0.2.ELsmp (RHEL 4.0) as well.
				 
				             total       used       free
shared    buffers     cached
				Mem:       4044496    4005516      38980
0      34108    2236636
				-/+ buffers/cache:    1734772    2309724
				Swap:      2097144     648244    1448900
				 
				If you've uncovered anything since
posting this message, please pass it along?
				 
				
________________________________

				From: ocfs2-users-bounces at oss.oracle.com
[mailto:ocfs2-users-bounces at oss.oracle.com] On Behalf Of Luis Freitas
				Sent: Friday, February 23, 2007 5:32 PM
				To: ocfs2-users at oss.oracle.com
				Subject: [Ocfs2-users] High on buffers
and deep on swap
				 
				Hi,
				 
				   This is a bit off topic, hope there
is not a problem.
				 
				   Anyone out there experiencing high
swapping with the kernel retaining a large amount of buffers? This used
to be a problem on 2.4, and I usually changed /proc/sys/vm/freepages to
fix it. But on 2.6 this parameter no longer exists.
				 
				    One of the servers here is holding
over 3.5Gb of cache even when using over 700Mb of swap, and free memory
is always low.
				 
				[oracle at br001sv0432 ~]$ free
				             total       used       free
shared    buffers     cached
				Mem:       5190736    4810880     379856
0     143032    3583868
				-/+ buffers/cache:    1083980    4106756
				Swap:      2048248     723064    1325184
				[oracle at br001sv0432 ~]$
				   I am tuning /proc/sys/vm/swappiness,
but this seems to have no effect at all. Changed from 60 to 10 and seems
to have no effect. The server runs Oracle RAC with OCFS2.
				 
				Regards,
				Luis
				 
				 
				  
				
________________________________

				Food fight?
<http://answers.yahoo.com/dir/index;_ylc=X3oDMTFvbGNhMGE3BF9TAzM5NjU0NTE
wOARfcwMzOTY1NDUxMDMEc2VjA21haWxfdGFnbGluZQRzbGsDbWFpbF90YWcx?link=ask&s
id=396545367>  Enjoy some healthy debate
				in the Yahoo! Answers Food & Drink Q&A.
<http://answers.yahoo.com/dir/index;_ylc=X3oDMTFvbGNhMGE3BF9TAzM5NjU0NTE
wOARfcwMzOTY1NDUxMDMEc2VjA21haWxfdGFnbGluZQRzbGsDbWFpbF90YWcx?link=ask&s
id=396545367> 
________________________________

	
_______________________________________________
				Ocfs2-users mailing list
				Ocfs2-users at oss.oracle.com
	
http://oss.oracle.com/mailman/listinfo/ocfs2-users


________________________________

				Be a PS3 game guru.
				Get your game face on with the latest
PS3 news and previews at Yahoo! Games.
<http://us.rd.yahoo.com/evt=49936/*http://videogames.yahoo.com> 


________________________________

		Finding fabulous fares is fun.
		Let Yahoo! FareChase search your favorite travel sites
<http://farechase.yahoo.com/promo-generic-14795097;_ylc=X3oDMTFtNW45amVp
BF9TAzk3NDA3NTg5BF9zAzI3MTk0ODEEcG9zAzEEc2VjA21haWx0YWdsaW5lBHNsawNxMS0w
Nw-->  to find flight and hotel bargains.


________________________________

Bored stiff? <http://us.rd.yahoo.com/evt=49935/*http://games.yahoo.com>
Loosen up...
Download and play hundreds of games for free
<http://us.rd.yahoo.com/evt=49935/*http://games.yahoo.com>  on Yahoo!
Games.


SEEBURGER AG 
Headquarters:
Edisonstraße 1 
D-75015 Bretten 
Tel.: 0 72 52/96-0 
Fax: 0 72 52/96-2222 
Internet: http://www.seeburger.de 
e-mail: info at seeburger.de 

Vorstand:
Bernd Seeburger, Axel Haas, Michael Kleeberg

Vorsitzender des Aufsichtsrats:
Dr. Franz Scherer

Handelsregister:
HRB 240708 Mannheim
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://oss.oracle.com/pipermail/ocfs2-users/attachments/20070410/e44677b2/attachment-0001.html


More information about the Ocfs2-users mailing list