FAT Performance Optimizations

[Gedare Bloom]

Very impressive numbers! If I read right, on your test the PC->device
transfer rate went from about 646 KB/sec to 3834 KB/sec, and device->PC
went from 571 KB/sec to 5652 KB/sec


On Thu, Dec 6, 2012 at 10:59 AM, Ralf Kirchner <
ralf.kirchner at embedded-brains.de> wrote:

> Hi Users,
> I would like to share some good news.
> We have been successful in eliminating a few performance bottlenecks in
> the FAT file system. The FAT file system now writes by cluster instead
> of writing by sector. Also reading from the device during a file write
> has been minimized.
> As a result the data rate during a file write went up while the CPU
> usage (for example of the USB sub-system) went down.
>
> Here are some test results. The test setup is:
> A simple application loaded to the MPC8313ERDB evaluation board. The
> test application provides an ftp server and formats a USB memory stick
> with a FAT file system (with 64 sectors per cluster). Then it mounts the
> USB memory stick to /work/0 on the target.
> Then from a PC the program curl gets used to transfer data from the PC
> to the USB memory at the target respectively from the USB memory stick
> to the PC.
>
> Previous status (without the performance optimization):
> PC -> MPC8313ERDB
> curl -T /dev/zero ftp://anonymous@192.168.96.93/work/0/zero
> %    Total   % Received Xferd Average Speed  Time    Time    Current
>                               Dload   Upload Total   Spent   Speed
> 100 78.8M    0 0        78.8M 0       646k  --:--:-- 0:02:04 438k
>
> MPC8313ERDB -> PC:
> curl -o /dev/null ftp://anonymous@192.168.96.93/work/0/zero
> %   Total %   Received Xferd Average Speed  Time    Time Current
>                              Dload Upload Total   Spent  Speed
> 100 80.7M 100 80.7M    0     571k      0 0:02:24 0:02:24 567k
>
>
> With FAT optimizations:
> PC -> MPC8313ERDB
> curl -T /dev/zero ftp://anonymous@192.168.96.93/work/0/zero
> %   Total    % Received Xferd Average Speed  Time     Time    Current
>                               Dload   Upload Total    Spent   Speed
> 100 1290M    0 0        1290M 0       3834k  --:--:-- 0:05:44 3914k
>
> [/work/0] # blkstats /dev/umass-sim-0-0
>
> -------------------------------------------------------------------------------
>                                DEVICE STATISTICS
>
> ----------------------+--------------------------------------------------------
>  READ HITS            | 96202
>  READ MISSES          | 328
>  READ AHEAD TRANSFERS | 118
>  READ BLOCKS          | 788
>  READ ERRORS          | 0
>  WRITE TRANSFERS      | 1179
>  WRITE BLOCKS         | 18539
>  WRITE ERRORS         | 0
>
> ----------------------+--------------------------------------------------------
>
> MPC8313ERDB -> PC:
> curl -o /dev/null  ftp://anonymous@192.168.96.93/work/0/zero
>   % Total    % Received Xferd Average Speed Time    Time    Current
>                               Dload  Upload Total   Spent   Speed
> 100 1291M  100 1291M        0 5652k       0 0:03:53 0:03:53 5900k
>
> [/work/0] # blkstats /dev/umass-sim-0-0
>
> -------------------------------------------------------------------------------
>                                DEVICE STATISTICS
>
> ----------------------+--------------------------------------------------------
>  READ HITS            | 767190
>  READ MISSES          | 7
>  READ AHEAD TRANSFERS | 2663
>  READ BLOCKS          | 10663
>  READ ERRORS          | 0
>  WRITE TRANSFERS      | 1
>  WRITE BLOCKS         | 1
>  WRITE ERRORS         | 0
>
> ----------------------+--------------------------------------------------------
>
> The performance impact will vary depending mostly on your target system
> and on the configured number of sectors per cluster.
>
> Kind Regards
> --------------------------------------------
> Embedded Brains GmbH
> Ralf Kirchner          Dornierstr. 4
> D-82178 Puchheim       Germany
> email: ralf.kirchner at embedded-brains.de
> Phone: +49-89-18 94 741-17
> Fax:   +49-89-18 94 741-08
>
> Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.
>
>
> _______________________________________________
> rtems-users mailing list
> rtems-users at rtems.org
> http://www.rtems.org/mailman/listinfo/rtems-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/users/attachments/20121206/7edcfb10/attachment-0001.html>