Re: [perf-discuss] Re: ufs performance

2007-03-21 Thread Alexander . Gorshenev
On Wed, 21 Mar 2007, allen mathias wrote: Alex, yes adding in the -library=stlport4 option does help a lot. Will read the manual on why. http://developers.sun.com/sunstudio/articles/cmp_stlport_libCstd.html Dan, yes the lseeks are way high on the CC version of the binary. Have posted this

[perf-discuss] Re: ufs performance

2007-03-21 Thread allen mathias
Alex, yes adding in the -library=stlport4 option does help a lot. Will read the manual on why. Dan, yes the lseeks are way high on the CC version of the binary. Have posted this out on the studio forumn. thanks for your prompt comments as always. This message posted from opensolaris.org

Re: [perf-discuss] Re: ufs performance

2007-03-21 Thread Dan Price
On Thu 22 Mar 2007 at 08:44AM, [EMAIL PROTECTED] wrote: > >If you run your two variants under truss -c, you may see that under CC > >this seems to generate an excess of lseek's. I'm not sure why: > > This is the problem with the default C++ library, I believe. > You're right. That works great.

Re: [perf-discuss] Re: ufs performance

2007-03-21 Thread Alexander . Gorshenev
On Wed, 21 Mar 2007, Dan Price wrote: On Wed 21 Mar 2007 at 10:24PM, allen mathias wrote: Apologies for not posting this earlier. This is the source code. int main() { string line; ofstream fsmo; fsmo.open("temp.temp.ae"); ifstream fsm; fsm.open("temp.a

Re: [perf-discuss] Re: ufs performance

2007-03-21 Thread Alexander . Gorshenev
Try using the following flags to reclaim the performance -library=stlport4 -O4 Alexander On Wed, 21 Mar 2007, allen mathias wrote: Dan here are the truss outputs -bash-3.00$ truss -c ./fileio-fs-fs #compiled using CC syscall seconds calls errors _exit

Re: [perf-discuss] Re: ufs performance

2007-03-21 Thread Dan Price
On Wed 21 Mar 2007 at 10:24PM, allen mathias wrote: > Apologies for not posting this earlier. This is the source code. > > int main() > { > string line; > ofstream fsmo; > fsmo.open("temp.temp.ae"); > > ifstream fsm; > fsm.open("temp.ae"); > getline

[perf-discuss] Re: ufs performance

2007-03-21 Thread allen mathias
Dan here are the truss outputs -bash-3.00$ truss -c ./fileio-fs-fs #compiled using CC syscall seconds calls errors _exit.000 1 read .345 17198 write 13.249 100 open .030 9 1 clos

[perf-discuss] Re: ufs performance

2007-03-21 Thread allen mathias
Apologies for not posting this earlier. This is the source code. #include #include #include using namespace std; int main() { string line; ofstream fsmo; fsmo.open("temp.temp.ae"); ifstream fsm; fsm.open("temp.ae"); getline(fsm, line);

Re: [perf-discuss] ufs performance

2007-03-21 Thread Dan Price
On Wed 21 Mar 2007 at 09:37PM, allen mathias wrote: > Also I notice that if instead of using fstreams I use cin and 'cat' > the file into my test program the time values are in favor of the > Solaris system. > > ufs logging is set on all disks. > > Can anyone shed some light on this. Allen, I t

Re: [perf-discuss] ufs performance

2007-03-21 Thread Alexander . Gorshenev
Can we see the simple program? Alexander On Wed, 21 Mar 2007, allen mathias wrote: Hi, I am attempting to understand why my Solaris with a UFS filesystem is taking an extrodinarilary larger time to read files from disk using a simple c++ program as compared to a linux system. The program is

[perf-discuss] ufs performance

2007-03-21 Thread allen mathias
Hi, I am attempting to understand why my Solaris with a UFS filesystem is taking an extrodinarilary larger time to read files from disk using a simple c++ program as compared to a linux system. The program is simple C++ code that uses fstreams to read a million lines and write them to a file as