[Wien] Help with Operation System

[Peter Blaha]

Thanks for the data. Very interesting indeed.

Intel has increased not only the clock speed of the processor, but also optimized it otherwise:
Intel Core i7 920, 2.66 GHz   -->    Core i7-2600 3.40 GHz
            99                 -->        37  sec !!!!   much larger speedup as only "GHz"-change

However, the memory bus has not increased much: (4 jobs in parallel)
           104                 -->        62  sec
     (almost no loss                  large loss when parallelizing)           			


Such data (and tests) can teach you, how to run a calculation most efficiently
(k-parallism and/or setting OMP_NUM_THREAD depending on number of k-points and processors).

Am 17.06.2011 13:41, schrieb M. Koshino:
> Dear Prof. Peter Blaha,
>
> Thank you for your advice.
>
>> Thanks for the benchmark times. This seems to be a nice processor.
>> Unfortunately, some of them are not very clear.
>>
>> a) Please set:
>> export OMP_NUM_THREADS=1 (2 and 4)
>> run    x lapw1
>> and check the timings (it should go faster when using more threads). You can also
>> verify using the "top" command whether the lapw1 command uses 1,2 or 4 cores.
>
> The result of running "x lapw1" after changing the number of threads is as follows:
>
> export OMP_NUM_THREADS=1
> test_case.output1:       TIME HAMILT (CPU)  =     6.8, HNS =     4.5, HORB =     0.0, DIAG =    26.2
> test_case.output1:       TIME HAMILT (WALL) =     6.9, HNS =     4.5, HORB =     0.0, DIAG =    26.4
> ("top" command shows cpu % = 99)
>
> export OMP_NUM_THREADS=2
> test_case.output1:       TIME HAMILT (CPU)  =    10.6, HNS =     5.7, HORB =     0.0, DIAG =    32.1
> test_case.output1:       TIME HAMILT (WALL) =     7.1, HNS =     2.9, HORB =     0.0, DIAG =    16.3
> ("top" command shows cpu % = 199)
>
> export OMP_NUM_THREADS=4
> test_case.output1:       TIME HAMILT (CPU)  =    20.1, HNS =     8.4, HORB =     0.0, DIAG =    48.9
> test_case.output1:       TIME HAMILT (WALL) =     7.2, HNS =     2.1, HORB =     0.0, DIAG =    12.7
> ("top" command shows cpu % = 398)
>
>
>>
>> b) x lapw1 -p    is not enough. What you should do is:
>>      create 4 identical directories, then open 4 windows, change into these 4 directories and
>>      run  "x lapw1" at the same time 2 or 4 times and check the timings.
>
> Running two x lapw1 on two different windows gives the results as follows:
>
> #1
> test_case.output1:       TIME HAMILT (CPU)  =     7.0, HNS =     4.7, HORB =     0.0, DIAG =    31.3
> test_case.output1:       TIME HAMILT (WALL) =     7.1, HNS =     4.7, HORB =     0.0, DIAG =    31.5
> #2
> test_case.output1:       TIME HAMILT (CPU)  =     7.0, HNS =     4.7, HORB =     0.0, DIAG =    31.1
> test_case.output1:       TIME HAMILT (WALL) =     7.1, HNS =     4.7, HORB =     0.0, DIAG =    31.3
>
>
> Running four x lapw1 on four different windows gives the results as follows:
>
> #1
> test_case.output1:       TIME HAMILT (CPU)  =     7.5, HNS =     5.3, HORB =     0.0, DIAG =    48.4
> test_case.output1:       TIME HAMILT (WALL) =     7.7, HNS =     5.4, HORB =     0.0, DIAG =    48.8
> #2
> test_case.output1:       TIME HAMILT (CPU)  =     7.6, HNS =     5.2, HORB =     0.0, DIAG =    48.4
> test_case.output1:       TIME HAMILT (WALL) =     7.8, HNS =     5.3, HORB =     0.0, DIAG =    48.8
> #3
> test_case.output1:       TIME HAMILT (CPU)  =     7.6, HNS =     5.2, HORB =     0.0, DIAG =    47.9
> test_case.output1:       TIME HAMILT (WALL) =     7.8, HNS =     5.3, HORB =     0.0, DIAG =    48.3
> #4
> test_case.output1:       TIME HAMILT (CPU)  =     7.8, HNS =     5.2, HORB =     0.0, DIAG =    48.0
> test_case.output1:       TIME HAMILT (WALL) =     8.0, HNS =     5.3, HORB =     0.0, DIAG =    48.4
>
>
>>
>> c) Most likely, the mpi-benchmark is not using mpi,.... But forget this test.
>>
>
> OK, I will skip this.
>
> Thank you again for the comment.
>
> FYI, I got a personal email from a person watching this mailing list but who was reluctant to write in English.
>
> Another option of OS for core i7 2600 could be RedHat Enterprise Linux 6.
>
> CentOS 6 can be another option which will be released next week.
>
> Scientific Linux 6 which is equivalent to CentOS is already released.
>
> Kind regards,
>
> Masanori Koshino
>
> Nanotube Research Center
>
> AIST, JAPAN
> m-koshino at aist.go.jp
>
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien

-- 

                                       P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------