[Cbe-oss-dev] spu-top

[Yury Serdyuk]

>>>spu-top: Context View
>>>Cpu(s) load avg: 0.09, 0.13, 0.22
>>>Spu(s) load avg: 3.43, 1.60, 1.18
>>>Cpu(s):  0.1%us,  0.3%sys,  0.2%wait,  0.0%nice, 99.4%idle
>>>Spu(s): 49.9%us,  0.1%sys,  0.0%wait, 50.0%idle
>>>
>>>   PID   TID USERNAME   S F  %SPU SPE     TIME BINARY
>>> 24429 24443 user002        U     0.2        4   14.230    mono
>>> 24429 24442 user002        U     0.2        5   14.231    mono
>>> 24429 24441 user002        U     0.2        6   14.232    mono
>>> 24429 24440 user002        U     0.2        7   14.232    mono
>>> 24429 24439 user002        L     0.0       -1   14.305    mono
>>> 24429 24438 user002        L     0.0       -1   14.305    mono
>>> 24429 24437 user002        L     0.0       -1   14.306    mono
>>> 24429 24436 user002        L     0.0       -1   14.308    mono
>>> 24429 24435 user002        U     0.2        0   14.342    mono
>>> 24429 24434 user002        U     0.2        1   14.343    mono
>>> 24429 24433 user002        U     0.2        2   14.345    mono
>>> 24429 24432 user002        U     0.2        3   14.347    mono 
>>>      
>>>
>>In fact,  real workload  is 100% for each SPU (except processes with "-1"),
>>and a total workload ( for 8 SPUs) is near 50 %:
>>
>>    
>>
>>>Spu(s): 49.9%us,  0.1%sys,  0.0%wait, 50.0%idle
>>>      
>>>
>
>That still sounds reasonable, you probably do a lot of context switches.
>
I think it's not true due to the following:
at first, I have tried a well known matrix multiplication program
(http://tu-dresden.de/die_tu_dresden/zentrale_einrichtungen/zih/forschung/architektur_und_leistungsanalyse_von_hochleistungsrechnern/cell/matmul/index_html/document_view?body_language=en)
on  the boxes  with  the different SDKs.

Here are the results:
1) QS22 blade, Fedora Core 7, SDK 3.0

> $ uname -a
> Linux cell8i 2.6.22-5.20070920bsc #1 SMP Tue Sep 25 10:49:16 CEST 2007 
> ppc64 ppc64 ppc64 GNU/Linux

An application run:

> @cell8i matmul]$ ./matmul -m 6144 -s 1

spu-top during the run:

> spu-top: Context View
> Cpu(s) load avg:   0.3%,  0.1%,  0.0%
> spu-top: Context View
> Cpu(s) load avg:   0.4%,  0.1%,  0.0%
> Spu(s) load avg:  20.2%,  9.8%,  3.8%
> Cpu(s): 24.9%us,  0.4%sys,  0.0%wait,  0.0%nice, 74.7%idle
> Spu(s):  6.2%us,  0.0%sys,  0.0%wait, 93.8%idle
>
>    PID   TID USERNAME   S F  %SPU SPE     TIME BINARY
>  30078 30079 user002    U   100.0   7    3.234 matmul

2) QS22 blade, Fedora Core 9, SDK 3.1:

> $ uname -a
> Linux cell8i-3 2.6.25.14-108.20080910bsc.ppc64 #1 SMP Fri Sep 12 
> 11:44:36 CEST 2008 ppc64 ppc64 ppc64 GNU/Linux

The same application run:

> cell8i-3 matmul]$ ./matmul -m 6144 -s 1

spu-top:

> spu-top: Context View
> Cpu(s) load avg: 0.56, 0.18, 0.06
> Spu(s) load avg: 0.10, 0.03, 0.01
> Cpu(s): 16.4%us,  0.3%sys,  0.0%wait,  0.0%nice, 83.3%idle
> Spu(s):  6.2%us,  0.0%sys,  0.0%wait, 93.8%idle
>
>    PID   TID USERNAME   S F  %SPU SPE     TIME BINARY
>  12887 12888 user002    U     0.0   7    4.328 matmul

Thanks.

Yury

PS 
There are also some kernel issues  which are present  on  QS blades in 
contrast
with PlayStation 3 (please see the following letter ).



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ozlabs.org/pipermail/cbe-oss-dev/attachments/20081121/52ab73ce/attachment.htm>