Re: fs_mark test on btrfs on 3.16.0-rc6+ #1 SMP

[Liu Bo <bo.li.liu@oracle.com>]

On Mon, Aug 18, 2014 at 05:38:17PM +0000, Ming Lei wrote:
> 
> Hi,
> 
> I ran the fs_mark test on a single empty hard drive. After the test, the df -h results are:
> 
> /dev/sdk1             917G   39G  832G   5% /ext4
> /dev/sdj1             932G   53G  850G   6% /btrfs
> 
> The test result for btrfs shows it ran 15 hours. Note there is no file/dir remove operation which I knew very slow compared with ext4.
> 
> [root@sh679 ~]# date;/root/fs_mark -v -n 1000000 -s 4096 -k -S 1 -D 1000 -N 1000 -d /btrfs/ -t 10;date
> Mon Aug 11 11:32:54 PDT 2014
> 
> #  /root/fs_mark  -v  -n  1000000  -s  4096  -k  -S  1  -D  1000  -N  1000  -d  /btrfs/  -t  10 
> #             Version 3.3, 10 thread(s) starting at Mon Aug 11 11:32:54 2014
> #             Sync method: INBAND FSYNC: fsync() per file in write loop.
> #             Directories:  Round Robin between directories across 1000 subdirectories with 1000 files per subdirectory.
> #             File names: 40 bytes long, (16 initial bytes of time stamp with 24 random bytes at end of name)
> #             Files info: size 4096 bytes, written with an IO size of 16384 bytes per write
> #             App overhead is time in microseconds spent in the test not doing file writing related system calls.
> #             All system call times are reported in microseconds
> FSUse%        Count         Size    Files/sec     App Overhead        CREAT (Min/Avg/Max)        WRITE (Min/Avg/Max)        FSYNC (Min/Avg/Max)         SYNC (Min/Avg/Max)        CLOSE (Min/Avg/Max)       UNLINK (Min/Avg/Max)
>      8     10000000         4096        184.0        155517800       33      372    93743        7       16     3094    16450    54015  5420340        0        0        0        1        4     7777        0        0        0
> Tue Aug 12 02:40:01 PDT 2014
> 
> For hours, the disk utilization was around 95% and cpu utilization for all 12 cores was very low and only one core showed around 26%wa.
> 
> 
> To compare with Ext4:
> The test for ext4 on a same model of hard drive ran 2.5 hours.  
> 
> [root@sh679 ~]# date;/root/fs_mark -v -n 1000000 -s 4096 -k -S 1 -D 1000 -N 1000 -d /ext4/ -t 10;date
> Fri Aug  8 17:13:56 PDT 2014
> #  /root/fs_mark  -v  -n  1000000  -s  4096  -k  -S  1  -D  1000  -N  1000  -d  /ext4/  -t  10 
> #             Version 3.3, 10 thread(s) starting at Fri Aug  8 17:13:56 2014
> #             Sync method: INBAND FSYNC: fsync() per file in write loop.
> #             Directories:  Round Robin between directories across 1000 subdirectories with 1000 files per subdirectory.
> #             File names: 40 bytes long, (16 initial bytes of time stamp with 24 random bytes at end of name)
> #             Files info: size 4096 bytes, written with an IO size of 16384 bytes per write
> #             App overhead is time in microseconds spent in the test not doing file writing related system calls.
> #             All system call times are reported in microseconds.
> 
> FSUse%        Count         Size    Files/sec     App Overhead        CREAT (Min/Avg/Max)        WRITE (Min/Avg/Max)        FSYNC (Min/Avg/Max)         SYNC (Min/Avg/Max)        CLOSE (Min/Avg/Max)       UNLINK (Min/Avg/Max)
>      9     10000000         4096        105.0        156950153       19      449  1741759        6       15  2069984    32368    94751  2044364        0        0        0        1        4     4149        0        0        0
> Sat Aug  9 19:41:14 PDT 2014
> 
> Is it a known issue with btrfs or do I need to adjust the default parameters for btrfs (I remember use the default to make btrfs)? 
> 
> Mount command shows:
> /dev/sdk1 on /ext4 type ext4 (rw,relatime,seclabel,data=ordered)
> /dev/sdj1 on /btrfs type btrfs (rw,relatime,seclabel,nospace_cache)

I guess it has something to do with nospace_cache, mount with -o space_cache
is the default option on my box.

Would you please give it a shot?

thanks,
-liubo

> 
> Thanks
> Ming
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html