2015.01.01 20:24 "Re: [Tiff] Fast TIFF Reading on Windows", by Aaron Boxer

2015.01.05 03:39 "Re: [Tiff] Fast TIFF Reading on Windows", by Aaron Boxer

Another finding: reading files with four threads (I have a quad core i7) is about twice as fast as single threaded reads.

On Sun, Jan 4, 2015 at 10:36 PM, Aaron Boxer <boxerab@gmail.com> wrote:

Gentlemen,

My very informal benchmark (windows 7, SSD, RamMAP tool to clear OS file cache before each run):

  1. twiddling windows flags made no difference to speed
  2. TIFFOpenfile with mode "ru" i.e. disabling memory mapping, was 3 times

slower than with default flag "r"

So, my conclusion is that memory mapping is the way to go for performance.

On Sat, Jan 3, 2015 at 12:24 AM, Edward Lam <edward@sidefx.com> wrote:

On Thu, January 1, 2015 15:20, Aaron Boxer wrote:

Thanks. Yes, my use case is: first-read of many 10-15 MB RGB

>> uncompressed

files. Unfortunately, I don't have control over how the files are stored, i.e. block size. One nice feature on Linux is fadvise, where I can specify FADV_SEQUENTIAL | FADV_DONTNEED, which would be right for my use case.

On Windows, the CreateFile() API has additional flags that you may wish to explore like FILE_FLAG_SEQUENTIAL_SCAN and/or FILE_FLAG_NO_BUFFERING

http://msdn.microsoft.com/en-us/library/windows/desktop/aa363858%28v=vs.85%29.aspx#caching_behavior