[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: EXT3 with 2.2.20 - Is it stable enough for a production server th at is used 24X7?



On Nov 15, 2001  13:04 +0100, Giulio Orsero wrote:
> Every time I run bonnie it starts killing processes as seen here:
> https://listman.redhat.com/pipermail/ext3-users/2001-July/000964.html
> I was told that 2.2.x VM needs some adjustements to handle journaling:
> https://listman.redhat.com/pipermail/ext3-users/2001-August/001002.html
> 
> Is it just me having this kind of problems? Or maybe Red Hat errata has
> some patch that clashes with ext3?
> I don't see anything in the 2.2.20 changelog that could solve my
> problems.
> This machine has 64MB mem.

How big is your journal?  If you have a very busy filesystem (which bonnie
would create) then potentially ALL of the journal is locking pages in RAM,
which would cause you to run out of available memory.  So, if you have a
32MB journal on an fs, and you are running bonnie there, it would be as if
you only have a 32MB RAM machine.  Take away memory for journals on other
filesystems, inode/dentry/buffer/page cache (which will be very large also
because of bonnie), and you quickly run out of available RAM.  Also, the
journaling mode you are using strongly affects how much RAM is locked by
the journal.  If you are using data=journal mode, then almost certainly
the journal will always be full.

Note that unless you have a real need for it, you can easily run with a
smaller journal.  The journal size is picked based on the size of the
filesystem, but it is somewhat arbitrary, since we don't even have tools
to check how full the journal is at a given time.  A long time ago, I
did a comparison with bonnie for default and minimum sized journals.

The results boil down to:
1) ext2 and ext3 are about the same on block I/O performance (expected)
2) ext2 is faster than ext3 on metadata intensive operations (expected)
3) ordered and writeback mode are about the same performance in most cases,
   so you may as well use ordered, because it is "safer"
4) decreasing the size of the journal doesn't impact data=ordered performance,
   because we don't have enough metadata to fill the 32MB journal
5) using data=journal mode hurts performance, because we write all the data
   to the journal and the disk, doubling I/O

What I'm missing here is the test with a 32MB journal for data=journal.  It
will obviously still be slower than data=ordered, but the question is how
much the small journal impacts the performance.  We can see for block writes
the data=journal performance is half of data=ordered, perfectly expected.

Cheers, Andreas


All tests run on 128MB RAM system, with X+xterms running, but not much else.
hdparm -t /dev/hda on 2.2.18-1.1  gives: 64 MB in  9.76 seconds = 6.56 MB/sec

Version  1.00       ------Sequential Output------ --Sequential Input- --Random-
2.2.18-0.3egcs      -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
LVM-0.9          MB K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
ext2            256            4198  64  1924  40            4120  14  21.9 1
320MB 4k fs         ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files:max:min        /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
     7:65536:512/64   113  47   101  16   562   1   113 61    43   7   417   1
Lynx,256,,,4198,64,1924,40,,,4120,14,21.9,1,7:65536:512/64,113,47,101,16,562,1,113,61,43,7,417,1

Version  1.00       ------Sequential Output------ --Sequential Input- --Random-
2.2.18-0.3egcs      -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
LVM-0.9          MB K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
ext3-0.0.5d     256            4150  47  1846  37            4104  14  21.3   1
320MB 4k fs         ------Sequential Create------ --------Random Create--------
32MB writeback jrnl -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files:max:min        /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
     7:65536:512/64    85  31   100  15   350   2    81  25    44   7   405   2
Lynx,256,,,4150,47,1846,37,,,4104,14,21.3,1,7:65536:512/64,85,31,100,15,350,2,81,25,44,7,405,2

Version  1.00       ------Sequential Output------ --Sequential Input- --Random-
2.2.18-egcs         -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
LVM-0.9          MB K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
ext3-0.0.5d     256            4168  58  1898  39            4123  16  21.3   1
320MB 4k fs         ------Sequential Create------ --------Random Create--------
32MB order journal  -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files:max:min        /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
     7:65536:512/64    82  22   104  16   362   2    82  28    43   6   410 3
Lynx,256,,,4168,58,1898,39,,,4123,16,21.3,1,7:65536:512/64,82,22,104,16,362,2,82,28,43,6,410,3

Version  1.00       ------Sequential Output------ --Sequential Input- --Random-
2.2.18-0.3egcs      -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
LVM-0.9          MB K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
ext3-0.0.5d     256            3988  59  1952  41            4105  13  21.5   1
320MB 4k fs         ------Sequential Create------ --------Random Create--------
4MB order journal   -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files:max:min        /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
     7:65536:512/64    81  23   103  16   354   2    79  26    43   6   124 2
Lynx,256,,,3988,59,1952,41,,,4105,13,21.5,1,7:65536:512/64,81,23,103,16,354,2,79,26,43,6,124,2

Version  1.00       ------Sequential Output------ --Sequential Input- --Random-
2.2.18-0.3egcs      -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
LVM-0.9          MB K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
ext3-0.0.5d     256            1877   4  1250  22            4153  14  15.3   1
320MB 4k fs         ------Sequential Create------ --------Random Create--------
4MB data journal    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files:max:min        /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
     7:65536:512/64    44   5   108  17   347   2    47   4    44   7   121   1
Lynx,256,,,1877,4,1250,22,,,4153,14,15.3,1,7:65536:512/64,44,5,108,17,347,2,47,4,44,7,121,1
--
Andreas Dilger
http://sourceforge.net/projects/ext2resize/
http://www-mddsp.enel.ucalgary.ca/People/adilger/





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]