This is the mail archive of the
mailing list for the Cygwin project.
Re: 1.5.20-1: ls -F slowness
Scott Roland wrote:
On Thursday, Nov 16 Corinna wrote:
I looked through this and I'm just puzzled. I'd expect 1.5.22 to be
as quick as 1.5.19 with respect to opening files on a share. And it
is in my tests. The number of usecs above is the time it takes to
convert an ASCII filename to a UNICODE filename, setting a couple of
flags and then call NtCreateFile. There are no changes in that code
between 1.5.19 and 1.5.20 which would in any way explain the slowness
you see. In my local tests it's always roughly about 1000 usecs.
Thanks for looking into this.
I did some monitoring of the performance problem and noticed that the
problems with 1.5.20 and newer are also effected by the performance of
the SMB server being accessed. Around lunch when no one is doing much
work it seems to be okay, but during normal works hour it can take 7
minutes. The server also has a virus scanner running on it, and when we
turn off the virus scanner the newer versions complete in a normal
amount of time.
So if the SMB server is not being used much then the ls -F has no
problems, unfortunately this is not the normal situation. I still don't
have any clue as to what the 1.5.20 or newer versions would be doing
special that make them slower than the 1.5.19 and earlier versions. (Are
the newer versions parallelizing the system calls and our server ends up
thrashing its cache?)
The first step would be to narrow down what call or calls are taking the
time when things slow down for you. A first approximation of this process is
to use 'strace'. You can look at the timestamp to figure out what's delaying
the access. Or you can just debug it. From there, you'll need to look at
what strategies may help. Here, a comparison of new and old "ways" may be
Does turning on and off 'smbntsec' make any difference?
Larry Hall http://www.rfk.com
RFK Partners, Inc. (508) 893-9779 - RFK Office
216 Dalton Rd. (508) 893-9889 - FAX
Holliston, MA 01746
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
Problem reports: http://cygwin.com/problems.html