Mail Archives: cygwin/1998/04/10/09:20:45
>From: Kevin Hughes <kh AT wg DOT icl DOT co DOT uk>
>Reply-To: "kh AT wg DOT icl DOT co DOT uk" <kh AT wg DOT icl DOT co DOT uk>
>To: "'Earnie Boyd'" <earnie_boyd AT hotmail DOT com>
>Cc: "gnu-win32 AT cygnus DOT com" <gnu-win32 AT cygnus DOT com>
>Subject: RE: Cygwin Performance - especially on a network
>Date: Wed, 8 Apr 1998 13:34:47 +0100
>
>Earnie,
> sorry about the long delay but I have been refining the measures etc.
and
>doing other work.
That's ok, I've been to busy to notice any delay, except for the mailing
list delay.
>
>The latest set of results for the open/fstat/close sequence are as
follows:
>(where times are in milliseconds (I hope))
>
>
>Local file access
>
> Min "Typical" Max
>Gnu/Cygnus 2.8 3.0 4.4
>Gnu/NT 0.73 0.8 1.7
>VC++ 0.75 0.8 2.85
Do you mean "Mingw32" when you say "Gnu/NT"? If so, I can believe it
based on packages I've ported to Mingw32.
>
>Remote file access
>
> Min "Typical Max
>Gnu/Cygnus 10.8 11.0 18.6
>Gnu/NT 3.4 3.5 5.0
>VC++ 3.4 3.5 6.0
>
>I have also looked at the file accesses per open/fstat/close
Did you catch the "Memory Leak" post concerning open/close? 16 bytes of
memory lost per issue of open/close pairs.
>
> Local Remote
>Gnu/Cygnus 20 23
>Gnu/NT 5 7
>VC++ 5 7
>
>The cost and accesses ration perfectly - the cost is all down to the
extra
>accesses that the cygnus layer issues.
>
Sounds correct to me.
>I guess this is all to do with the emulation of the Unix filestore
>semantics but I do not understand why. I would have thought that a
simple
>stat to determine if its a link and then an open would be sufficient -
but
>I am very naive in this.
>
>Any information much appreciated.
Have a look at the documentation for Win32 stat and fstat. Inode is
returned as 0, so cygwin32 emulates it. There is no true "link" on
Win32 and lstat isn't supplied in the libraries by MS. File privilege
bit masks have to be emulated to the extent of getting a true UNIX
filesystem emulation. And, the directory has to be mapped against the
cygwin32 mount table to find the file. I'm sure there are other things
going on to; but, don't know what as yet.
>
>Kevin
>
>PS - I have read that NtOpenFile bypasses the win32 layer and allows
mixed
>case file names to be passed.
>
I think that Sergey has already incorporated Mixed Case on NTFS into the
cygwin32.dll based on earlier posts.
- \\||//
---o0O0--Earnie--0O0o----
-earnie_boyd AT hotmail DOT com-
------ooo0O--O0ooo-------
PS: Did you try your test on a FAT file system?
______________________________________________________
Get Your Private, Free Email at http://www.hotmail.com
-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request AT cygnus DOT com" with one line of text: "help".
- Raw text -