HD to SSD question.

[Mihai T. Lazarescu]

On Tue, Aug 20, 2013 at 04:02:42PM +0200, Heinz Diehl wrote:

> On 20.08.2013, Heinz Diehl wrote: 
> 
> > Then, open another one and run fsync-tester. The numbers that count to
> > compare different elevators on your system is the output fsync-tester
> > generates while your machine is generating the "bigfile".
> 
> And while your're at it, you could also consider doing some
> fsmark runs (without the load), e.g.:
> 
>  ./fs_mark -S 1 -D 10000 -N 100000 -d /home/htd/fsmark/test -s 65536 -t4 -w 4096 -F 
> 
> Notice the -t switch, which lets you specify the number of
> threads used.

Here they are:

CFQ:

    #  ./fs_mark  -S  1  -D  10000  -N  100000  -d  /home/mihai/tmp/c/fs_mark-3.3  -s  65536  -t4  -w  4096  -F 
    # Version 3.3, 4 thread(s) starting at Tue Aug 20 23:30:40 2013
    # Sync method: INBAND FSYNC: fsync() per file in write loop.
    # Directories: Round Robin between directories across 10000
    # subdirectories with 100000 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 65536 bytes, written with an IO size of
    # 4096 bytes per write
    # App overhead is time in microseconds spent in the test not
    # doing file writing related system calls.
    #
    FSUse%        Count         Size    Files/sec     App Overhead
        77         4000        65536         50.4            88160
        77         8000        65536         43.2           104150
        77        12000        65536         44.4            94512

Deadline:

    #  ./fs_mark  -S  1  -D  10000  -N  100000  -d  /home/mihai/tmp/c/fs_mark-3.3/test  -s  65536  -t4  -w  4096  -F 
    # Version 3.3, 4 thread(s) starting at Tue Aug 20 23:42:19 2013
    # Sync method: INBAND FSYNC: fsync() per file in write loop.
    # Directories: Round Robin between directories across 10000
    # subdirectories with 100000 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 65536 bytes, written with an IO size of
    # 4096 bytes per write
    # App overhead is time in microseconds spent in the test not
    # doing file writing related system calls.
    #
    FSUse%        Count         Size    Files/sec     App Overhead
        77         4000        65536         47.6            92902
        77         8000        65536         41.6           101888
        77        12000        65536         39.6            97937

Noop:

    #  ./fs_mark  -S  1  -D  10000  -N  100000  -d  /home/mihai/tmp/c/fs_mark-3.3/test  -s  65536  -t4  -w  4096  -F 
    # Version 3.3, 4 thread(s) starting at Tue Aug 20 23:48:10 2013
    # Sync method: INBAND FSYNC: fsync() per file in write loop.
    # Directories: Round Robin between directories across 10000
    # subdirectories with 100000 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 65536 bytes, written with an IO size of
    # 4096 bytes per write
    # App overhead is time in microseconds spent in the test not
    # doing file writing related system calls.
    #
    FSUse%        Count         Size    Files/sec     App Overhead
        77         4000        65536         46.8            89478
        77         8000        65536         42.0           101337
        77        12000        65536         43.2            95834

It does not look like significant differences between schedulers.

Mihai