Re: dynamic allocation of swap disk space

Re: dynamic allocation of swap disk space

[Ray Bryant]

Silviu Marin-Caea wrote:
>
> 
> My desktop has been thrashing the disk for a couple of hours because
> the swap space was exhausted.  And I have the ambition to leave it alone
> to see if it ever comes out of the thrashing.  Of course, it's not usable
> at all during this time, I'm writing this on the laptop.
> 
>

You've got a couple problems mixed together here.

(1)  Swap space fills up because you have overcommitted memory.  In principle, 
filling up swap space has nothing to do with "thrashing".
(2)  "thrashing" is a characteristic of a poorly performing program in a
demand paging virtual memory system typically caused by trying to run a 
program with a resident size that is smaller than required.  Systems can 
thrash without filling up swap space.  It is true that systems can thrash AND 
fill up swap space, but it is not always so.

You either need (1)  more main memory, (2) reduce the number of programs you 
are running simultaneously, or (3) better behaving programs, or all three. 
Increasing the amount of swap space will just use more disk.  It won't cause 
your system to stop thrashing since that is driven by what is going on in 
memory, not what is going on on the disk.

Anyway, if swap space is full, then main memory is (nearly) full, and this is 
not a good time to do major surgery to your file system (e. g. increase the 
size of the swap space).  You may not be able to allocate the required memory 
to complete that job and then your system will crash.
> -
> 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/

Re: dynamic allocation of swap disk space

[John Bradford]

Quote from Ray Bryant <raybry@sgi.com>:
> 
> 
> Silviu Marin-Caea wrote:
> >
> > 
> > My desktop has been thrashing the disk for a couple of hours because
> > the swap space was exhausted.  And I have the ambition to leave it alone
> > to see if it ever comes out of the thrashing.  Of course, it's not usable
> > at all during this time, I'm writing this on the laptop.
> > 
> >
> 
> You've got a couple problems mixed together here.
> 
> (1)  Swap space fills up because you have overcommitted memory.  In principle, 
> filling up swap space has nothing to do with "thrashing".
> (2)  "thrashing" is a characteristic of a poorly performing program in a
> demand paging virtual memory system typically caused by trying to run a 
> program with a resident size that is smaller than required.  Systems can 
> thrash without filling up swap space.  It is true that systems can thrash AND 
> fill up swap space, but it is not always so.
> 
> You either need (1)  more main memory, (2) reduce the number of programs you 
> are running simultaneously, or (3) better behaving programs, or all three. 
> Increasing the amount of swap space will just use more disk.  It won't cause 
> your system to stop thrashing since that is driven by what is going on in 
> memory, not what is going on on the disk.

Not necessarily.  Increasing swap can allow more physical RAM to be used for
caching data from disk.

Imagine a system with limited physical RAM, and limited swap space, running a
process which causes a lot of filesystem activity on the same physical disk
as is being used for swap.  If the total RAM, both physical and swap is almost
completely full, increasing the swap space may allow some data from physical
RAM to be swapped out, in favour of caching filesystem data from the disk.

Without knowing more details of the original poster's machine, it's difficult
to give specific advice about how to solve the problem.

John.

Re: dynamic allocation of swap disk space

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 601 bytes --]

On Tue, 11 May 2004 16:52:15 BST, John Bradford said:

> Imagine a system with limited physical RAM, and limited swap space, running a
> process which causes a lot of filesystem activity on the same physical disk
> as is being used for swap.  If the total RAM, both physical and swap is almost
> completely full, increasing the swap space may allow some data from physical
> RAM to be swapped out, in favour of caching filesystem data from the disk.

Possible, but wouldn't that imply that the value of /proc/sys/vm/swappiness
is very mis-set and causing a too-high estimate of the working set size?


[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: dynamic allocation of swap disk space

[John Bradford]

Quote from Valdis.Kletnieks@vt.edu:
> 
> On Tue, 11 May 2004 16:52:15 BST, John Bradford said:
> 
> > Imagine a system with limited physical RAM, and limited swap space, running a
> > process which causes a lot of filesystem activity on the same physical disk
> > as is being used for swap.  If the total RAM, both physical and swap is almost
> > completely full, increasing the swap space may allow some data from physical
> > RAM to be swapped out, in favour of caching filesystem data from the disk.
> 
> Possible, but wouldn't that imply that the value of /proc/sys/vm/swappiness
> is very mis-set and causing a too-high estimate of the working set size?

Yes - to be honest, I was thinking of very low spec systems, where swap may
well be less than or equal to physical RAM size.

John.

Re: dynamic allocation of swap disk space

[Ray Bryant]

Hi John,

John Bradford wrote:
>
> 
> Not necessarily.  Increasing swap can allow more physical RAM to be used for
> caching data from disk.
> 
> Imagine a system with limited physical RAM, and limited swap space, running a
> process which causes a lot of filesystem activity on the same physical disk
> as is being used for swap.  If the total RAM, both physical and swap is almost
> completely full, increasing the swap space may allow some data from physical
> RAM to be swapped out, in favour of caching filesystem data from the disk.

Hmmm... Lets see, we have a program (or set of programs) in memory that is 
thrashing, i. e. it is page faulting at a rate that is higher than the vm 
system can supply pages, so it is spending its time waiting for pages and the 
disk subsystem is busy.  Now, if we increase the amount of data cached from 
disk, without increasing main memory, we've decreased the amount of memory 
available to the thrashing program, perhaps making its problems worse?

Well, I guess all this shows is that with the vm subsystem, speculation is 
often useless, one has to fire up the box with a carefully constructed 
workload and see what happens.

> 
> Without knowing more details of the original poster's machine, it's difficult
> to give specific advice about how to solve the problem.
>

I certainly agree with that.  Time for this thread to die, I think.  :-)

Cheers.

> John.
> 

-- 
Best Regards,
Ray
-----------------------------------------------
                   Ray Bryant
512-453-9679 (work)         512-507-7807 (cell)
raybry@sgi.com             raybry@austin.rr.com
The box said: "Requires Windows 98 or better",
            so I installed Linux.
-----------------------------------------------

Re: dynamic allocation of swap disk space

[Silviu Marin-Caea]

John Bradford wrote:
> Quote from Ray Bryant <raybry@sgi.com>:
> 
>>
>>Silviu Marin-Caea wrote:
>>
>>>
>>>My desktop has been thrashing the disk for a couple of hours because
>>>the swap space was exhausted.  And I have the ambition to leave it alone
>>>to see if it ever comes out of the thrashing.  Of course, it's not usable
>>>at all during this time, I'm writing this on the laptop.
>>>
>>>
>>
>>You've got a couple problems mixed together here.
>>
>>(1)  Swap space fills up because you have overcommitted memory.  In principle, 
>>filling up swap space has nothing to do with "thrashing".
>>(2)  "thrashing" is a characteristic of a poorly performing program in a
>>demand paging virtual memory system typically caused by trying to run a 
>>program with a resident size that is smaller than required.  Systems can 
>>thrash without filling up swap space.  It is true that systems can thrash AND 
>>fill up swap space, but it is not always so.
>>
>>You either need (1)  more main memory, (2) reduce the number of programs you 
>>are running simultaneously, or (3) better behaving programs, or all three. 
>>Increasing the amount of swap space will just use more disk.  It won't cause 
>>your system to stop thrashing since that is driven by what is going on in 
>>memory, not what is going on on the disk.
> 
> 
> Not necessarily.  Increasing swap can allow more physical RAM to be used for
> caching data from disk.
> 
> Imagine a system with limited physical RAM, and limited swap space, running a
> process which causes a lot of filesystem activity on the same physical disk
> as is being used for swap.  If the total RAM, both physical and swap is almost
> completely full, increasing the swap space may allow some data from physical
> RAM to be swapped out, in favour of caching filesystem data from the disk.
> 
> Without knowing more details of the original poster's machine, it's difficult
> to give specific advice about how to solve the problem.

1 GB RAM desktop machine, with 32 MB swap, kernel 2.6.3.  In normal 
operation, the swap is 80% free.  It started the thrashing when I loaded 
a HUGE page (generated from a database) in konqueror.

The issue is not to solve _my_ problem, I really don't care about it 
that much, the issue would be to solve this sort of problem for 
everyone.  I've had disk thrashing with filled up swap on a server 
(kernel 2.4) and that was much worse.  Anyway, I'll start using swapd 
and see what happens.  Some time in the future, I'll post here the 
conclusion.

Re: dynamic allocation of swap disk space

[Oliver Neukum]

> thrash without filling up swap space.  It is true that systems can thrash AND 
> fill up swap space, but it is not always so.

However the system will also thrash if you have inactive dirty pages without
file backing taking up to much main memory. Albeit it is more a theoretical
possibilty.

	Regards
		Oliver

dynamic allocation of swap disk space

[Silviu Marin-Caea]

It's the same old whine.  Please include
http://sourceforge.net/projects/swapd/ or
http://sourceforge.net/projects/dynswapd/
or whatever in the official kernel.

My desktop has been thrashing the disk for a couple of hours because
the swap space was exhausted.  And I have the ambition to leave it alone
to see if it ever comes out of the thrashing.  Of course, it's not usable
at all during this time, I'm writing this on the laptop.

This is the only serious bug (I consider it a bug) in the linux kernel
that I encountered on occasions.  And it was not fixed in 2.6.  Forgot to
mention, my desktop is running 2.6.3.

The way I see the solution is: allocate swap space dynamically, until
there is no need for more or the disk becomes nearly full.  If that
happens, then start thrashing it, all right.  Then when the condition is
gone and things are back to normal deallocate the additional swap.

Old whine or not, it's a real problem.  Thanks.

PS: I'm not subscribed.

Re: dynamic allocation of swap disk space

[John Bradford]

Hi,

Quote from "Silviu Marin-Caea" <silviu@genesys.ro>:
> The way I see the solution is: allocate swap space dynamically, until
> there is no need for more or the disk becomes nearly full.  If that
> happens, then start thrashing it, all right.  Then when the condition is
> gone and things are back to normal deallocate the additional swap.

Very bad idea in my opinion.

Over allocating swap space is a BAD practice, but the effects are usually not
as apparent as under allocating swap space, and such a system works most of
the time, albeit often not as efficiently as it could do.  The solution is to
allocate the correct amount of swap space.  How much is that?  There is not
really a simple answer, and it's certainly not as simple as twice the physical
RAM.

A run-away process on a server with too much swap may well cause the machine
to become very unresponsive, whereas if the amount of virtual memory available
had been little more than what was expected to be required, the run-away
process would have terminated, allowing the problem to be fixed with minimal
disruption to other services.

John.

Re: dynamic allocation of swap disk space

[Silviu Marin-Caea]

John Bradford wrote:

>>The way I see the solution is: allocate swap space dynamically, until
>>there is no need for more or the disk becomes nearly full.  If that
>>happens, then start thrashing it, all right.  Then when the condition is
>>gone and things are back to normal deallocate the additional swap.
> 
> 
> Very bad idea in my opinion.

Most likely quite so, I'm not a guru yet :-)

> Over allocating swap space is a BAD practice, but the effects are usually not

How about dynamically allocating up to a certain limit.

Say, you have 256 MB or even less (to save space), and you allocate when 
needed up to 1 GB, then stop allocating, thrash disk, let the kernel 
detect and kill the runaway process.

Re: dynamic allocation of swap disk space

[Paul Jackson]

> How about dynamically allocating up to a certain limit.

Given that disk space is essentially "free", how does that
differ from simply allocating swap files up to the desired
limit size?

And having a bigger swap space wouldn't help your thrashing
problem.  You're not thrashing (unusable system, mucho disk
activity) for lack of swap space.  You're thrashing because
you are trying to use the swap space you already have for
active pages.

-- 
                          I won't rest till it's the best ...
                          Programmer, Linux Scalability
                          Paul Jackson <pj@sgi.com> 1.650.933.1373