forked from StephenDennis/gpu-load-sensor
-
Notifications
You must be signed in to change notification settings - Fork 0
/
README
94 lines (85 loc) · 4 KB
/
README
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
About
-----
This package contains a Grid Engine Load Sensor for both Cuda and OpenCL.
Author
------
Stephen Dennis, [email protected]
License
-------
This package is licensed under GPL. See COPYING
Use
---
need qsub examples!
Building and Installing
-----------------------
-Package assumes you have installed cuda in /usr/local/cuda which is default.
-If you have installed else where you may need to edit the Makefile include path
-cuda-sensor uses dlopen() to load its libraries. You may need to either set LD_LIBRARY_PATH
or update your ld.so.conf to allow a successful load of the load sensor
-Run cuda-sensor directly and press return to confirm that it can start and detect
your cuda configuration.
-running 'cuda-sensor -help' will display a script you can use to setup Grid Engine
for use with the cuda-sensor.
-Add the cuda_sensor to your Grid Engine using either 'qconf -msconf' or 'qconf -msconf hostname'
in the following form.
[sd@head ~]$ qconf -sconf zorg
#zorg.farmotron.org:
mailer /bin/mail
xterm /usr/bin/X11/xterm
load_sensor /home/sd/gpu-load-sensor/cuda_sensor
-Confirm that the load sensor has loaded as follows
sd@zorg:~$ ps axf|grep -e sge -e gpu |grep -v grep
11855 ? Sl 0:02 /opt/sge/bin/lx-amd64/sge_execd
12067 ? S 0:00 \_ /bin/sh -c /home/sd/gpu-load-sensor/cuda_sensor
12068 ? S 0:00 \_ /home/sd/gpu-load-sensor/cuda_sensor
-Confirm load_values are available with
sd@zorg:~$ qconf -se zorg
hostname zorg.farmotron.org
load_scaling NONE
complex_values NONE
load_values arch=lx-amd64,num_proc=2,mem_total=3837.156250M, \
swap_total=0.000000M,virtual_total=3837.156250M, \
load_avg=0.010000,load_short=0.000000, \
load_medium=0.010000,load_long=0.000000, \
mem_free=2526.710938M,swap_free=0.000000M, \
virtual_free=2526.710938M,mem_used=1310.445312M, \
swap_used=0.000000M,virtual_used=1310.445312M, \
cpu=7.900000,m_topology=SCC,m_topology_inuse=SCC, \
m_socket=1,m_core=2,m_thread=2,np_load_avg=0.005000, \
np_load_short=0.000000,np_load_medium=0.005000, \
np_load_long=0.000000,cuda.count=1,cuda.error=0, \
cuda.strerror=No Error,cuda.0.name=GeForce 8200, \
cuda.0.major=1,cuda.0.minor=1, \
cuda.0.totalGlobalMem=131792896, \
cuda.0.sharedMemPerBlock=16384,cuda.0.regsPerBlock=8192, \
cuda.0.warpSize=32,cuda.0.memPitch=262144, \
cuda.0.maxThreadsPerBlock=512,cuda.0.maxThreadsDim=512, \
cuda.0.maxGridSize=65535,cuda.0.totalConstMem=65536, \
cuda.0.clockRate=1200000,cuda.0.textureAlignment=256, \
cuda.0.deviceOverlap=0,cuda.0.multiProcessorCount=1, \
cuda.0.kernelExecTimeoutEnabled=1,cuda.0.integrated=1, \
cuda.0.canMapHostMemory=1,cuda.0.computeNode=0
processors 2
user_lists NONE
xuser_lists NONE
projects NONE
xprojects NONE
usage_scaling NONE
report_variables NONE
TODO
----
-qsub examples
-Test against OSX
-Add a version string, including recording it as a load value
-It appears that cuda version number is not being expressed properly through
the load values
-Add a cuda consummable value which is updated from the load sensor
-some of the complex types need to be adjusted
Platforms Tested
----------------
April 15 2011
MF OK on GeForce 8400GS (512MB) on Ubuntu 10.10 (Maverick) Cuda 3.2.16
April 15 2011
SD OK on GeForce 8200 (1024MB) on Ubuntu 9.10 (Karmic) Cuda 2.2 SGE 6.2u5
April 15 2011
SD OK on GeForce 8200 (1024MB) on Ubuntu 10.04 (Lucid) Cuda 3.2.16 UGE 8.0.0