Re: vmalloc - kmalloc and page locks

Re: vmalloc - kmalloc and page locks

[Deas, Jim]

Thanks for the clarification. I am using mlockall on the user application. This 
still leaves me with the mystery of why my system usage goes from 3% to
50% randomly while playing data streams off the harddrive. I can also make
system usage stay at 50% by opening a third stream.
These streams are pulling data at 1.5MB/s each from different files (same HD).
I don't see that as a big strain on the hardware (4.5MB/s total data rate).
Where else should I look to find the bottleneck/latency issue?

Regards,
J. Deas


-----Original Message-----
From: Gábor Lénárt [mailto:lgb@lgb.hu]
Sent: Tuesday, July 22, 2003 6:13 AM
To: Deas, Jim
Cc: linux-kernel@vger.kernel.org
Subject: Re: vmalloc - kmalloc and page locks


Errrrr ... Sorry, I did not read your mail carefully ;-(
I meant in case of a user process you can use mlock() and such :)
AFAIK the kernel itself is not pagable ...

On Tue, Jul 22, 2003 at 03:07:18PM +0200, Gábor Lénárt wrote:
> Please read something about the mlock() and/or mlockall() functions.
> The prototype can be found in [/usr/include/]sys/mman.h
> You can read there:
> 
> /* Guarantee all whole pages mapped by the range [ADDR,ADDR+LEN) to
>    be memory resident.  */
> extern int mlock (__const void *__addr, size_t __len) __THROW;
> [...]
> /* Cause all currently mapped pages of the process to be memory resident
>    until unlocked by a call to the `munlockall', until the process exits,
>    or until the process calls `execve'.  */
> extern int mlockall (int __flags) __THROW;
> 
> On Tue, Jul 22, 2003 at 06:00:14AM -0700, Deas, Jim wrote:
> > How can I look at what memory are being paged out of memory in the kernel
> > or how to lock kmalloc and vmalloc pages so they do not get put to swap?
> [...]
> 
> - Gábor (larta'H)
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

-- 
- Gábor (larta'H)

Re: vmalloc - kmalloc and page locks

[Antonio Vargas]

On Tue, Jul 22, 2003 at 06:57:26AM -0700, Deas, Jim wrote:
> Thanks for the clarification. I am using mlockall on the user application. This 
> still leaves me with the mystery of why my system usage goes from 3% to
> 50% randomly while playing data streams off the harddrive. I can also make
> system usage stay at 50% by opening a third stream.
> These streams are pulling data at 1.5MB/s each from different files (same HD).
> I don't see that as a big strain on the hardware (4.5MB/s total data rate).
> Where else should I look to find the bottleneck/latency issue?
> 

Are you doing read() or mmap()?

If you are only reading and not modifying, having it mmap'ed as read only
and sequential access could be much faster.

Late 2.4 and 2.5+ kernels do read-ahead both with read() and mmap() access.

For extra effect, try using madvise() to ask the kernel to prefetch data
and to discard the already used ones.


> 
> -----Original Message-----
> From: G?bor L?n?rt [mailto:lgb@lgb.hu]
> Sent: Tuesday, July 22, 2003 6:13 AM
> To: Deas, Jim
> Cc: linux-kernel@vger.kernel.org
> Subject: Re: vmalloc - kmalloc and page locks
> 
> 
> Errrrr ... Sorry, I did not read your mail carefully ;-(
> I meant in case of a user process you can use mlock() and such :)
> AFAIK the kernel itself is not pagable ...
> 
> On Tue, Jul 22, 2003 at 03:07:18PM +0200, G?bor L?n?rt wrote:
> > Please read something about the mlock() and/or mlockall() functions.
> > The prototype can be found in [/usr/include/]sys/mman.h
> > You can read there:
> > 
> > /* Guarantee all whole pages mapped by the range [ADDR,ADDR+LEN) to
> >    be memory resident.  */
> > extern int mlock (__const void *__addr, size_t __len) __THROW;
> > [...]
> > /* Cause all currently mapped pages of the process to be memory resident
> >    until unlocked by a call to the `munlockall', until the process exits,
> >    or until the process calls `execve'.  */
> > extern int mlockall (int __flags) __THROW;
> > 
> > On Tue, Jul 22, 2003 at 06:00:14AM -0700, Deas, Jim wrote:
> > > How can I look at what memory are being paged out of memory in the kernel
> > > or how to lock kmalloc and vmalloc pages so they do not get put to swap?
> > [...]
> > 
> > - G?bor (larta'H)
> > -
> > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > Please read the FAQ at  http://www.tux.org/lkml/
> 
> -- 
> - G?bor (larta'H)
> 
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

-- 

1. Dado un programa, siempre tiene al menos un fallo.
2. Dadas varias lineas de codigo, siempre se pueden acortar a menos lineas.
3. Por induccion, todos los programas se pueden
   reducir a una linea que no funciona.

Re: vmalloc - kmalloc and page locks

[Gábor Lénárt]

Errrrr ... Sorry, I did not read your mail carefully ;-(
I meant in case of a user process you can use mlock() and such :)
AFAIK the kernel itself is not pagable ...

On Tue, Jul 22, 2003 at 03:07:18PM +0200, Gábor Lénárt wrote:
> Please read something about the mlock() and/or mlockall() functions.
> The prototype can be found in [/usr/include/]sys/mman.h
> You can read there:
> 
> /* Guarantee all whole pages mapped by the range [ADDR,ADDR+LEN) to
>    be memory resident.  */
> extern int mlock (__const void *__addr, size_t __len) __THROW;
> [...]
> /* Cause all currently mapped pages of the process to be memory resident
>    until unlocked by a call to the `munlockall', until the process exits,
>    or until the process calls `execve'.  */
> extern int mlockall (int __flags) __THROW;
> 
> On Tue, Jul 22, 2003 at 06:00:14AM -0700, Deas, Jim wrote:
> > How can I look at what memory are being paged out of memory in the kernel
> > or how to lock kmalloc and vmalloc pages so they do not get put to swap?
> [...]
> 
> - Gábor (larta'H)
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

-- 
- Gábor (larta'H)

Re: vmalloc - kmalloc and page locks

[William Lee Irwin III]

On Tue, Jul 22, 2003 at 06:00:14AM -0700, Deas, Jim wrote:
> How can I look at what memory are being paged out of memory in the kernel
> or how to lock kmalloc and vmalloc pages so they do not get put to swap?
>  I have a program that runs great 90% of the time but the other 10%
> of the time the system usage (using 'top')goes from 3% to 50% and latency goes out
> the window!  I am assuming this is due to some of my buffers getting swaped 
> out as it often corrects itself and runs well the majority of time.
> Doubling the base memory from 256M to 512M did nothing to fix this.
> I need some way to find out who is holding up the process.
> Any suggestions? linux-newbe did not give me any replys, if
> this is the wrong groups can someone redirect me?

Linux is not a pageable kernel; neither vmalloc() nor kmalloc() return
swappable memory.


-- wli

Re: vmalloc - kmalloc and page locks

[Gábor Lénárt]

Please read something about the mlock() and/or mlockall() functions.
The prototype can be found in [/usr/include/]sys/mman.h
You can read there:

/* Guarantee all whole pages mapped by the range [ADDR,ADDR+LEN) to
   be memory resident.  */
extern int mlock (__const void *__addr, size_t __len) __THROW;
[...]
/* Cause all currently mapped pages of the process to be memory resident
   until unlocked by a call to the `munlockall', until the process exits,
   or until the process calls `execve'.  */
extern int mlockall (int __flags) __THROW;

On Tue, Jul 22, 2003 at 06:00:14AM -0700, Deas, Jim wrote:
> How can I look at what memory are being paged out of memory in the kernel
> or how to lock kmalloc and vmalloc pages so they do not get put to swap?
[...]

- Gábor (larta'H)

vmalloc - kmalloc and page locks

[Deas, Jim]

How can I look at what memory are being paged out of memory in the kernel
or how to lock kmalloc and vmalloc pages so they do not get put to swap?
 I have a program that runs great 90% of the time but the other 10%
of the time the system usage (using 'top')goes from 3% to 50% and latency goes out
the window!  I am assuming this is due to some of my buffers getting swaped 
out as it often corrects itself and runs well the majority of time.
Doubling the base memory from 256M to 512M did nothing to fix this.
I need some way to find out who is holding up the process.
Any suggestions? linux-newbe did not give me any replys, if
this is the wrong groups can someone redirect me?

Best Regards,
J. Deas