Data/File Performance¶

The display modules have several different options when it comes to storing files/data:

As the display modules comes with 16Mb external QSPI flash and the LPC4088 MCU has 512Kb of internal flash those memories can be used without adding any extra costs. If the storage requirements are greater then external removable memories (USB MemorySticks or uSD cards) can be bought and plugged in.

Reading¶

The following summary shows the read performance of each of the different memories. Files of different sizes are read into a large buffer in external SDRAM and the total time is recorded. For comparison the files are also available as arrays in internal flash (IFLASH[]) and external QSPI flash (QSPI[]).

----
Summary:

  File Information
          Filename       Size
          --------       ----
     bench_600.bmp        623 bytes
     bench_10k.bmp       9703 bytes
    bench_100k.bmp     129040 bytes
    bench_350k.bmp     375959 bytes
      bench_1m.bmp     920339 bytes
      bench_4m.bmp    3701897 bytes

  Read times (in us)
          Filename   uSD Card        USB    QSPI FS   IFLASH[]     QSPI[]
          --------   --------        ---    -------   --------     ------
     bench_600.bmp       1998      23916       1862          9         40
     bench_10k.bmp       8340     125333       2554        118        656
    bench_100k.bmp      96314    1535869      11652       3101       8696
    bench_350k.bmp     268215    4427903      30466        N/A      22780
      bench_1m.bmp     654990   10799233      71937        N/A      65376
      bench_4m.bmp    2633862   43409273     283842        N/A     224272

  Read speeds
          Filename  uSD Card     USB          QSPI FS      IFLASH[]     QSPI[]
          --------  ------------ ------------ ------------ ------------ ------------
     bench_600.bmp   304.50 Kb/s   25.44 Kb/s  326.74 Kb/s   66.02 Mb/s   14.85 Mb/s
     bench_10k.bmp  1136.16 Kb/s   75.60 Kb/s 3710.10 Kb/s   78.42 Mb/s   14.11 Mb/s
    bench_100k.bmp  1308.38 Kb/s   82.05 Kb/s   10.56 Mb/s   39.68 Mb/s   14.15 Mb/s
    bench_350k.bmp  1368.86 Kb/s   82.92 Kb/s   11.77 Mb/s         N/A    15.74 Mb/s
      bench_1m.bmp  1372.19 Kb/s   83.23 Kb/s   12.20 Mb/s         N/A    13.43 Mb/s
      bench_4m.bmp  1372.56 Kb/s   83.28 Kb/s   12.44 Mb/s         N/A    15.74 Mb/s

---

Some notes:

1. The IFLASH[] and QSPI[] columns show direct memcpy from that memory to the destination buffer and has no overhead of a file system or by being read in small chunks as is done in a FAT file system.
2. The lower performance of the small files is likely due to the fact that the file system overhead for opening a file is constant and independent of the size of the file. So smaller files get a relatively larger overhead.
3. The USB read times are very low with the USBHost stack in mbed. The performance is almost identical on the original 1768 mbed platform. A test (not show here) of the non-mbed libusb stack available in [[|LPCOpen]] shows that the libusb stack is 1.5 - 5.5 times faster.

Writing¶

The following summary shows the write performance of each of the different memories. The source data is read into external SDRAM and then written to the destination file. Only the file operations are timed.

----
Summary:

  File Information
          Filename       Size
          --------       ----
     bench_600.bmp        623 bytes
     bench_10k.bmp       9703 bytes
    bench_100k.bmp     129040 bytes
    bench_350k.bmp     375959 bytes
      bench_1m.bmp     920339 bytes
      bench_4m.bmp    3701897 bytes

  Write times (in us)
          Filename   uSD Card        USB    QSPI FS
          --------   --------        ---    -------
     bench_600.bmp      15136     400059      55723
     bench_10k.bmp      26183     185175     114838
    bench_100k.bmp      90345    1917807    1082966
    bench_350k.bmp     250364    5298330    3399382
      bench_1m.bmp     591573   13282119    6368029
      bench_4m.bmp    2182433   53461098   25471267

  Write speeds
          Filename  uSD Card     USB          QSPI FS
          --------  ------------ ------------ ------------
     bench_600.bmp    40.20 Kb/s 1557.27 b/s    10.92 Kb/s
     bench_10k.bmp   361.90 Kb/s   51.17 Kb/s   82.51 Kb/s
    bench_100k.bmp  1394.83 Kb/s   65.71 Kb/s  116.36 Kb/s
    bench_350k.bmp  1466.45 Kb/s   69.29 Kb/s  108.00 Kb/s
      bench_1m.bmp  1519.29 Kb/s   67.67 Kb/s  141.14 Kb/s
      bench_4m.bmp  1656.47 Kb/s   67.62 Kb/s  141.93 Kb/s

---

Some notes:

1. The uSD card is by far the fastest medium to write to.
2. QSPI flash is slow to write to but that is expected. The strength of the QSPI flash is in the fast read speeds.
3. The USB write times are a bit slower than the USB read times.
4. The lower performance of the small files is likely due to the fact that the file system overhead for opening a file is constant and independent of the size of the file. So smaller files get a relatively larger overhead.

Software¶

The program used to do the tests can be downloaded here:

The program has two large files image_data.c and image_data_qspi.c which are placed in internal flash and external QSPI flash respectively. The image_data files contains the test files as c arrays like this:

const unsigned char img_bench_100k[] IMAGE_LOCATION = {
  0xFF, 0xD8, 0xFF, 0xE0, 0x00, 0x10, 0x4A, 0x46, 0x49, 0x46, 0x00, 0x01, 0x01, 0x01, 0x01, 0x2C, 0x01
  ...
  0x28, 0x00, 0xA2, 0x8A, 0x28, 0x00, 0xA2, 0x8A, 0x28, 0x00, 0xA2, 0x8A, 0x28, 0x03, 0xFF, 0xD9, 0x00
};

const unsigned int img_size_bench_100k IMAGE_LOCATION = sizeof(img_bench_100k);

The img2c.py script was used to convert the test files into arrays.

In runtime the program starts by saving the data in the arrays as files on each of the three supported file systems. When all file systems are prepared each file/array is read and timed and the result is then summarized as shown above.

After the read tests the user is asked to press the USER button on the display module to run the write tests. This is to allow running the read tests multiple times without constantly having to rewrite the file system (as that could degrade it over time). Before the write tests the existing files (if any) are deleted and the QSPI file system is reformatted so that no performance is gained by writing the same file to the same spot again.