«

Hi,
I'm intending to use multiprocessing on a freebsd machine (6.3
release, quad core, 8cpus, amd64). I see in the doc that on this
platform I can't use synchronize:

ImportError: This platform lacks a functioning sem_open
implementation, therefore, the required synchronization primitives
needed will not function, see issue 3770.

As far as I can tell, I have no need to synchronize the processes--I
have several processes run separately and I need to know when they're
all finished; there's no communication between them and each owns its
own log file for output.

Is anyone using multiprocessing on FreeBSD and run into any other
gotchas?
thanks,
--Tim Arnold

On Mar 2, 2010, at 11:31 AM, Tim Arnold wrote:

> Hi,
> I'm intending to use multiprocessing on a freebsd machine (6.3
> release, quad core, 8cpus, amd64). I see in the doc that on this
> platform I can't use synchronize:
>
> ImportError: This platform lacks a functioning sem_open
> implementation, therefore, the required synchronization primitives
> needed will not function, see issue 3770.
>
> As far as I can tell, I have no need to synchronize the processes--I
> have several processes run separately and I need to know when they're
> all finished; there's no communication between them and each owns its
> own log file for output.
>
> Is anyone using multiprocessing on FreeBSD and run into any other
> gotchas?

Hi Tim,
I don't use multiprocessing but I've written two low-level IPC
packages, one for SysV IPC and the other for POSIX IPC.

I think that multiprocessing prefers POSIX IPC (which is where
sem_open() comes from). I don't know what it uses if that's not
available, but SysV IPC seems a likely alternative. I must emphasize,
however, that that's a guess on my part.

FreeBSD didn't have POSIX IPC support until 7.0, and that was sort of
broken until 7.2. As it happens, I was testing my POSIX IPC code
against 7.2 last night and it works just fine.

SysV IPC works under FreeBSD 6 (and perhaps earlier versions; 6 is the
oldest I've tested). ISTR that by default each message queue is
limited to 2048 bytes in total size. 'sysctl kern.ipc' can probably
tell you that and may even let you change it. Other than that I can't
think of any SysV limitations that might bite you.

HTH
Philip

On Mar 2, 11:52*am, Philip Semanchuk <phi...@semanchuk.com> wrote:
> On Mar 2, 2010, at 11:31 AM, Tim Arnold wrote:
>
>
>
>
>
> > Hi,
> > I'm intending to use multiprocessing on a freebsd machine (6.3
> > release, quad core, 8cpus, amd64). I see in the doc that on this
> > platform I can't use synchronize:
>
> > ImportError: This platform lacks a functioning sem_open
> > implementation, therefore, the required synchronization primitives
> > needed will not function, see issue 3770.
>
> > As far as I can tell, I have no need to synchronize the processes--I
> > have several processes run separately and I need to know when they're
> > all finished; there's no communication between them and each owns its
> > own log file for output.
>
> > Is anyone using multiprocessing on FreeBSD and run into any other
> > gotchas?
>
> Hi Tim,
> I don't use multiprocessing but I've written two low-level IPC *
> packages, one for SysV IPC and the other for POSIX IPC.
>
> I think that multiprocessing prefers POSIX IPC (which is where *
> sem_open() comes from). I don't know what it uses if that's not *
> available, but SysV IPC seems a likely alternative. I must emphasize, *
> however, that that's a guess on my part.
>
> FreeBSD didn't have POSIX IPC support until 7.0, and that was sort of *
> broken until 7.2. As it happens, I was testing my POSIX IPC code *
> against 7.2 last night and it works just fine.
>
> SysV IPC works under FreeBSD 6 (and perhaps earlier versions; 6 is the *
> oldest I've tested). ISTR that by default each message queue is *
> limited to 2048 bytes in total size. 'sysctl kern.ipc' can probably *
> tell you that and may even let you change it. Other than that I can't *
> think of any SysV limitations that might bite you.
>
> HTH
> Philip

Hi Philip,
Thanks for that information. I wish I could upgrade the machine to
7.2! alas, out of my power. I get the following results from sysctl:
% sysctl kern.ipc | grep msg
kern.ipc.msgseg: 2048
kern.ipc.msgssz: 8
kern.ipc.msgtql: 40
kern.ipc.msgmnb: 2048
kern.ipc.msgmni: 40
kern.ipc.msgmax: 16384

I'll write some test programs using multiprocessing and see how they
go before committing to rewrite my current code. I've also been
looking at 'parallel python' although it may have the same issues.
http://www.parallelpython.com/

thanks again,
--Tim

On Mar 2, 12:59*pm, Tim Arnold <a_j...@bellsouth.net> wrote:
> On Mar 2, 11:52*am, Philip Semanchuk <phi...@semanchuk.com> wrote:
> > On Mar 2, 2010, at 11:31 AM, Tim Arnold wrote:
>
> > > Hi,
> > > I'm intending to use multiprocessing on a freebsd machine (6.3
> > > release, quad core, 8cpus, amd64). I see in the doc that on this
> > > platform I can't use synchronize:
>
> > > ImportError: This platform lacks a functioning sem_open
> > > implementation, therefore, the required synchronization primitives
> > > needed will not function, see issue 3770.
>
> > > As far as I can tell, I have no need to synchronize the processes--I
> > > have several processes run separately and I need to know when they're
> > > all finished; there's no communication between them and each owns its
> > > own log file for output.
>
> > > Is anyone using multiprocessing on FreeBSD and run into any other
> > > gotchas?
>
> > Hi Tim,
> > I don't use multiprocessing but I've written two low-level IPC *
> > packages, one for SysV IPC and the other for POSIX IPC.
>
> > I think that multiprocessing prefers POSIX IPC (which is where *
> > sem_open() comes from). I don't know what it uses if that's not *
> > available, but SysV IPC seems a likely alternative. I must emphasize, *
> > however, that that's a guess on my part.
>
> > FreeBSD didn't have POSIX IPC support until 7.0, and that was sort of *
> > broken until 7.2. As it happens, I was testing my POSIX IPC code *
> > against 7.2 last night and it works just fine.
>
> > SysV IPC works under FreeBSD 6 (and perhaps earlier versions; 6 is the *
> > oldest I've tested). ISTR that by default each message queue is *
> > limited to 2048 bytes in total size. 'sysctl kern.ipc' can probably *
> > tell you that and may even let you change it. Other than that I can't *
> > think of any SysV limitations that might bite you.
>
> > HTH
> > Philip
>
> Hi Philip,
> Thanks for that information. I wish I could upgrade the machine to
> 7.2! alas, out of my power. *I get the following results from sysctl:
> % sysctl kern.ipc | grep msg
> kern.ipc.msgseg: 2048
> kern.ipc.msgssz: 8
> kern.ipc.msgtql: 40
> kern.ipc.msgmnb: 2048
> kern.ipc.msgmni: 40
> kern.ipc.msgmax: 16384
>
> I'll write some test programs using multiprocessing and see how they
> go before committing to rewrite my current code. I've also been
> looking at 'parallel python' although it may have the same issues.http://www.parallelpython.com/
>
> thanks again,
> --Tim

Well that didn't work out well. I can't import either Queue or Pool
from multiprocessing, so I'm back to the drawing board. I'll see now
how parallel python does on freebsd.

--Tim

On Mar 2, 2010, at 1:31 PM, Tim Arnold wrote:

> On Mar 2, 12:59 pm, Tim Arnold <a_j...@bellsouth.net> wrote:
>> On Mar 2, 11:52 am, Philip Semanchuk <phi...@semanchuk.com> wrote:
>>> On Mar 2, 2010, at 11:31 AM, Tim Arnold wrote:
>>
>>>> Hi,
>>>> I'm intending to use multiprocessing on a freebsd machine (6.3
>>>> release, quad core, 8cpus, amd64). I see in the doc that on this
>>>> platform I can't use synchronize:
>>
>>>> ImportError: This platform lacks a functioning sem_open
>>>> implementation, therefore, the required synchronization primitives
>>>> needed will not function, see issue 3770.
>>
>>>> As far as I can tell, I have no need to synchronize the
>>>> processes--I
>>>> have several processes run separately and I need to know when
>>>> they're
>>>> all finished; there's no communication between them and each owns
>>>> its
>>>> own log file for output.
>>
>>>> Is anyone using multiprocessing on FreeBSD and run into any other
>>>> gotchas?
>>
>>> Hi Tim,
>>> I don't use multiprocessing but I've written two low-level IPC
>>> packages, one for SysV IPC and the other for POSIX IPC.
>>
>>> I think that multiprocessing prefers POSIX IPC (which is where
>>> sem_open() comes from). I don't know what it uses if that's not
>>> available, but SysV IPC seems a likely alternative. I must
>>> emphasize,
>>> however, that that's a guess on my part.
>>
>>> FreeBSD didn't have POSIX IPC support until 7.0, and that was sort
>>> of
>>> broken until 7.2. As it happens, I was testing my POSIX IPC code
>>> against 7.2 last night and it works just fine.
>>
>>> SysV IPC works under FreeBSD 6 (and perhaps earlier versions; 6 is
>>> the
>>> oldest I've tested). ISTR that by default each message queue is
>>> limited to 2048 bytes in total size. 'sysctl kern.ipc' can probably
>>> tell you that and may even let you change it. Other than that I
>>> can't
>>> think of any SysV limitations that might bite you.
>>
>>> HTH
>>> Philip
>>
>> Hi Philip,
>> Thanks for that information. I wish I could upgrade the machine to
>> 7.2! alas, out of my power. I get the following results from sysctl:
>> % sysctl kern.ipc | grep msg
>> kern.ipc.msgseg: 2048
>> kern.ipc.msgssz: 8
>> kern.ipc.msgtql: 40
>> kern.ipc.msgmnb: 2048
>> kern.ipc.msgmni: 40
>> kern.ipc.msgmax: 16384
>>
>> I'll write some test programs using multiprocessing and see how they
>> go before committing to rewrite my current code. I've also been
>> looking at 'parallel python' although it may have the same issues.http://www.parallelpython.com/
>>
>> thanks again,
>> --Tim
>
> Well that didn't work out well. I can't import either Queue or Pool
> from multiprocessing, so I'm back to the drawing board. I'll see now
> how parallel python does on freebsd.


Sorry to hear that didn't work for you. Should you need to get down to
the nuts & bolts level, my module for SysV IPC is here:
http://semanchuk.com/philip/sysv_ipc/


Good luck with Parallel Python,
Philip

On 3/2/2010 12:59 PM, Tim Arnold wrote:
>
> I'll write some test programs using multiprocessing and see how they
> go before committing to rewrite my current code. I've also been
> looking at 'parallel python' although it may have the same issues.
> http://www.parallelpython.com/
>

parallelpython works for me on FreeBSD 6.2.