3.9 freeze

diego wrote:
>       UVM amap201783 39322K  39322K 39322K1237975710   
0     0 
>
16,32,64,128,256,512,1024,2048,4096,8192,16384,32768,65536

I have noticed that the above "UVM amap" HighUse
value is equal to the 
Limit value.

As I have already said, the PC of mine that occasional
freezes has high 
values of "VM amap" too.

So I'm asking again: what happens when the Limit value is
reached?
Can it make the pc to freeze?
What can we do to avoid this?


Thanks.

-- 
___________________________________________________
     __
    |-                      giannicineomedia.it
    |ederico Giannici      http://www.neomedia.it
___________________________________________________

On Thu, Jul 13, 2006 at 05:39:40PM +0200, Federico Giannici
wrote:
> I have noticed that the above "UVM amap"
HighUse value is equal to the 
> Limit value.

Indeed it looks suspicious. Not my area, though, so I'd
have to look at
the code to know the exact consequences. But yes, it's a
possibility.

Perhaps Mickey or Ted can enlighten us?

-p.

Federico,

Your diagnosis is correct, that freeze can be the result of
reaching the
limit for UVM amap allocations. These get used by the kernel
to describe
anonymous memory mappings, and mmap malloc() puts the UVM
subsystem
under a higher load of those, eventually reaching the limit.
Until an
appropriate solution is found, you can try bumping the
number of pages
in the kernel's memory map (NKMEMPAGES).

-p.

Federico,

Your diagnosis is correct, that freeze can be the result of
reaching the
limit for UVM amap allocations. These get used by the kernel
to describe
anonymous memory mappings, and mmap malloc() puts the UVM
subsystem
under a higher load of those, eventually reaching the limit.
Until an
appropriate solution is found, you can try bumping the
number of pages
in the kernel's memory map (NKMEMPAGES).

-p.

Pedro Martelletto wrote:
> Federico,
> 
> Your diagnosis is correct, that freeze can be the
result of reaching the
> limit for UVM amap allocations. These get used by the
kernel to describe
> anonymous memory mappings, and mmap malloc() puts the
UVM subsystem
> under a higher load of those, eventually reaching the
limit. Until an
> appropriate solution is found, you can try bumping the
number of pages
> in the kernel's memory map (NKMEMPAGES).

I'm not sure of what variables to set and where.
Is it correct to add the following line to the kernel
configuration file?

option  NKMEMPAGES_MAX  65536


Thanks.

-- 
___________________________________________________
     __
    |-                      giannicineomedia.it
    |ederico Giannici      http://www.neomedia.it
___________________________________________________

On Tue, Jul 18, 2006 at 07:34:00PM +0200, Federico Giannici
wrote:
> I'm not sure of what variables to set and where.

options(4) should tell you that.

-p.

Pedro Martelletto wrote:
> Federico,
> 
> Your diagnosis is correct, that freeze can be the
result of reaching the
> limit for UVM amap allocations. These get used by the
kernel to describe
> anonymous memory mappings, and mmap malloc() puts the
UVM subsystem
> under a higher load of those, eventually reaching the
limit. Until an
> appropriate solution is found, you can try bumping the
number of pages
> in the kernel's memory map (NKMEMPAGES).

I have just installed a new 3.9-stable kernel with doubled 
NKMEMPAGES_MAX (and maxusers).

I'll keep you informed if this solved the problem...

Bye.


P.S.
It seems to me that under amd64 the "UVM Amap"
usage is much higher then 
under i386. So, even if by default the limit is the double
of the i386, 
it seems not enough.

-- 
___________________________________________________
     __
    |-                      giannicineomedia.it
    |ederico Giannici      http://www.neomedia.it
___________________________________________________

Pedro Martelletto wrote:
> Federico,
> 
> Your diagnosis is correct, that freeze can be the
result of reaching the
> limit for UVM amap allocations. These get used by the
kernel to describe
> anonymous memory mappings, and mmap malloc() puts the
UVM subsystem
> under a higher load of those, eventually reaching the
limit. Until an
> appropriate solution is found, you can try bumping the
number of pages
> in the kernel's memory map (NKMEMPAGES).

I have just installed a new 3.9-stable kernel with doubled 
NKMEMPAGES_MAX (and maxusers).

I'll keep you informed if this solved the problem...

Bye.


P.S.
It seems to me that under amd64 the "UVM Amap"
usage is much higher then 
under i386. So, even if by default the limit is the double
of the i386, 
it seems not enough.

-- 
___________________________________________________
     __
    |-                      giannicineomedia.it
    |ederico Giannici      http://www.neomedia.it
___________________________________________________

Pedro Martelletto wrote:
> Federico,
> 
> Your diagnosis is correct, that freeze can be the
result of reaching the
> limit for UVM amap allocations. These get used by the
kernel to describe
> anonymous memory mappings, and mmap malloc() puts the
UVM subsystem
> under a higher load of those, eventually reaching the
limit. Until an
> appropriate solution is found, you can try bumping the
number of pages
> in the kernel's memory map (NKMEMPAGES).

I have just installed a new 3.9-stable kernel with doubled 
NKMEMPAGES_MAX (and maxusers).

I'll keep you informed if this solved the problem...

Bye.


P.S.
It seems to me that under amd64 the "UVM Amap"
usage is much higher then 
under i386. So, even if by default the limit is the double
of the i386, 
it seems not enough.

-- 
___________________________________________________
     __
    |-                      giannicineomedia.it
    |ederico Giannici      http://www.neomedia.it
___________________________________________________

Pedro Martelletto wrote:
> Federico,
> 
> Your diagnosis is correct, that freeze can be the
result of reaching the
> limit for UVM amap allocations. These get used by the
kernel to describe
> anonymous memory mappings, and mmap malloc() puts the
UVM subsystem
> under a higher load of those, eventually reaching the
limit. Until an
> appropriate solution is found, you can try bumping the
number of pages
> in the kernel's memory map (NKMEMPAGES).

I have just installed a new 3.9-stable kernel with doubled 
NKMEMPAGES_MAX (and maxusers).

I'll keep you informed if this solved the problem...

Bye.


P.S.
It seems to me that under amd64 the "UVM Amap"
usage is much higher then 
under i386. So, even if by default the limit is the double
of the i386, 
it seems not enough.

-- 
___________________________________________________
     __
    |-                      giannicineomedia.it
    |ederico Giannici      http://www.neomedia.it
___________________________________________________