questions on multiple thread programming

[lehe]

I perhaps have to resort to multiple thread programming to reduce computational time. I am now studying how to use POSIX library.
I can access a cluster with 8 CPUs, each of which has one core. I was wondering how many threads to create would get approximately best time performance on my cluster? Is it the same number of CPUs? If I am using multiple thread library like POSIX, will the library take care of which thread run on which CPU or do I have to specify this in my code?

Thanks!

[bithub]

To get the best performance, create one thread per core. The library (or more specifically, the OS scheduler) will take care of which thread runs on which core. You don't need to worry about this.

[vart]

Originally Posted by bithub

To get the best performance, create one thread per core. The library (or more specifically, the OS scheduler) will take care of which thread runs on which core. You don't need to worry about this.

I would say 1 worker thread per CPU

plus you can have several IO threads for example which will most of the time spend in waiting state while io operation is been completed

[bithub]

I would say 1 worker thread per CPU

Can you explain why you would use 1 thread per CPU instead of 1 thread per core? To software applications, there is no difference.

plus you can have several IO threads for example which will most of the time spend in waiting state while io operation is been completed

Why would you create several threads to wait for IO? For the best performance, async IO is the best way to go.

[vart]

Originally Posted by bithub

Can you explain why you would use 1 thread per CPU instead of 1 thread per core? To software applications, there is no difference.

Why would you create several threads to wait for IO? For the best performance, async IO is the best way to go.

by cpu I mean core

several IO? because you can have a logger thread writign logs to file,
db connection threads - requestion data from remote db server
sochet thread - processing some network connection etc...

it is too complicated to put all these task on one io thread

[Codeplug]

>> I can access a cluster with 8 CPUs, each of which has one core.
Multi-threading is for a single machine. For distributed-parallel programming you'll need something like MPI. Unless you're running a distributed OS providing SSI like openMosix.

gg

[lehe]

Thanks!
Yes I remember people called my server "cluster". But how to check if it is distributed system and running a distributed OS providing SSI? And which shall I use, multi-threading or distributed-parallel programming?

Here is the info of my server from uname:

Linux 2.6.9-78.0.8.ELsmp #1 SMP Wed Nov 19 19:42:38 EST 2008 x86_64 x86_64 x86_64 GNU/Linux

And also info from cat /proc/cpuinfo

processor : 0
vendor_id : GenuineIntel
cpu family : 15
model : 4
model name : Intel(R) Xeon(TM) MP CPU 3.66GHz
stepping : 1
cpu MHz : 3657.792
cache size : 1024 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 1
fpu : yes
fpu_exception : yes
cpuid level : 5
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx lm pni monitor ds_cpl est tm2 cid cx16 xtpr
bogomips : 7321.81
clflush size : 64
cache_alignment : 128
address sizes : 40 bits physical, 48 bits virtual
power management:

processor : 1
vendor_id : GenuineIntel
cpu family : 15
model : 4
model name : Intel(R) Xeon(TM) MP CPU 3.66GHz
stepping : 1
cpu MHz : 3657.792
cache size : 1024 KB
physical id : 4
siblings : 2
core id : 4
cpu cores : 1
fpu : yes
fpu_exception : yes
cpuid level : 5
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx lm pni monitor ds_cpl est tm2 cid cx16 xtpr
bogomips : 7315.15
clflush size : 64
cache_alignment : 128
address sizes : 40 bits physical, 48 bits virtual
power management:

processor : 2
vendor_id : GenuineIntel
cpu family : 15
model : 4
model name : Intel(R) Xeon(TM) MP CPU 3.66GHz
stepping : 1
cpu MHz : 3657.792
cache size : 1024 KB
physical id : 3
siblings : 2
core id : 3
cpu cores : 1
fpu : yes
fpu_exception : yes
cpuid level : 5
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx lm pni monitor ds_cpl est tm2 cid cx16 xtpr
bogomips : 7315.02
clflush size : 64
cache_alignment : 128
address sizes : 40 bits physical, 48 bits virtual
power management:

processor : 3
vendor_id : GenuineIntel
cpu family : 15
model : 4
model name : Intel(R) Xeon(TM) MP CPU 3.66GHz
stepping : 1
cpu MHz : 3657.792
cache size : 1024 KB
physical id : 7
siblings : 2
core id : 7
cpu cores : 1
fpu : yes
fpu_exception : yes
cpuid level : 5
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx lm pni monitor ds_cpl est tm2 cid cx16 xtpr
bogomips : 7315.10
clflush size : 64
cache_alignment : 128
address sizes : 40 bits physical, 48 bits virtual
power management:

processor : 4
vendor_id : GenuineIntel
cpu family : 15
model : 4
model name : Intel(R) Xeon(TM) MP CPU 3.66GHz
stepping : 1
cpu MHz : 3657.792
cache size : 1024 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 1
fpu : yes
fpu_exception : yes
cpuid level : 5
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx lm pni monitor ds_cpl est tm2 cid cx16 xtpr
bogomips : 7315.04
clflush size : 64
cache_alignment : 128
address sizes : 40 bits physical, 48 bits virtual
power management:

processor : 5
vendor_id : GenuineIntel
cpu family : 15
model : 4
model name : Intel(R) Xeon(TM) MP CPU 3.66GHz
stepping : 1
cpu MHz : 3657.792
cache size : 1024 KB
physical id : 4
siblings : 2
core id : 4
cpu cores : 1
fpu : yes
fpu_exception : yes
cpuid level : 5
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx lm pni monitor ds_cpl est tm2 cid cx16 xtpr
bogomips : 7314.70
clflush size : 64
cache_alignment : 128
address sizes : 40 bits physical, 48 bits virtual
power management:

processor : 6
vendor_id : GenuineIntel
cpu family : 15
model : 4
model name : Intel(R) Xeon(TM) MP CPU 3.66GHz
stepping : 1
cpu MHz : 3657.792
cache size : 1024 KB
physical id : 3
siblings : 2
core id : 3
cpu cores : 1
fpu : yes
fpu_exception : yes
cpuid level : 5
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx lm pni monitor ds_cpl est tm2 cid cx16 xtpr
bogomips : 7314.99
clflush size : 64
cache_alignment : 128
address sizes : 40 bits physical, 48 bits virtual
power management:

processor : 7
vendor_id : GenuineIntel
cpu family : 15
model : 4
model name : Intel(R) Xeon(TM) MP CPU 3.66GHz
stepping : 1
cpu MHz : 3657.792
cache size : 1024 KB
physical id : 7
siblings : 2
core id : 7
cpu cores : 1
fpu : yes
fpu_exception : yes
cpuid level : 5
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx lm pni monitor ds_cpl est tm2 cid cx16 xtpr
bogomips : 7315.06
clflush size : 64
cache_alignment : 128
address sizes : 40 bits physical, 48 bits virtual
power management:

[matsp]

That is one machine for all intents and purposes.

--
Mats

[lehe]

Hi Matsp,
Thanks!
From where did you tell it is not a distribution system? Does "cluster" only refer to distributed system?
So I should stick to multi-threading programming?

[matsp]

Originally Posted by lehe

Hi Matsp,
Thanks!
From where did you tell it is not a distribution system? Does "cluster" only refer to distributed system?
So I should stick to multi-threading programming?

A cluster/distributed system will not show that you have 8 cores in /proc/cpuinfo. It will only show the cores of THAT system, which will be one or two perhaps.

--
Mats

[Codeplug]

The term "cluster" implies more than one physical machine (PC). You have a single machine with 8 cores, so Posix "multi-threading" (with pthreads or processes) is appropriate in this case.

>> I was wondering how many threads to create would get approximately best time performance on my cluster?
The answer can depend on what you're doing, but a typical answer is one thread per core - as bithub and vart mentioned.

>> will the library take care of which thread run on which CPU or do I have to specify this in my code?
The typical answer is that you let your OS/library take care of this for you.

gg

[matsp]

Originally Posted by Codeplug

>> will the library take care of which thread run on which CPU or do I have to specify this in my code?
The typical answer is that you let your OS/library take care of this for you.

And unless you have SPECIFIC knowledge from understanding the problem nature, the OS in question and processors in question, it is nearly always a waste of time to try to improve on what the OS/library does.

--
Mats