Slab allocator . . . cache? WTF is it?

Slab allocator . . . cache? WTF is it?

[john moser]

Mem:    775616k total,   747740k used,    27876k free,    96584k buffers
Swap:   250480k total,    71340k used,   179140k free,   298852k cached


This is somewhat distressing.  Last time this happened, I started opening every
program I had (including every OpenOffice.org product and every browser I had
installed), and the "cached" value dropped quickly.

I'm wondering, what IS cache?  It seems to increase even when swap is not used,
and sometimes when there's no swap partition enabled.  It also seems to cause
me to run into swap when I have ample ram available, assuming that cache is just
some sort of cache that is copied from and mirrors another portion of ram for
some sort of speed increase.  It's wasteful to me, and I want to more understand
its implimentation and its purpose, and in all honesty limit its impact if possible.
I got this RAM upgrade because I was using about 676M of RAM total, including swap,
at peak; and now I find myself using 820M RAM at peak and about 750-800 continually.

As always, i'm not on the list, so I'd prefer to be CC'd replies.

_____________________________________________________________
Linux.Net -->Open Source to everyone
Powered by Linare Corporation
http://www.linare.com/

Re: Slab allocator . . . cache? WTF is it?

[Fox!MURDER]

Hi,
I'm kind of a kernel-dev newbie but I'll try to explain my understanding 
in BFU-friendly way.
The cache you're seeing here is a memory used to store IO requests and 
data when you request some operation on IO. For example if you are 
writing to hard drive then (if your partition is not mounted with sync 
option or its not some special case) the data you are writing are put in 
the cache and the disk starts writing. And because your memory is fast, 
let's say it can do a 500MiB/s, then your write is completed in 0.1sec 
and your program can continue doing other stuff. Data you requested to 
write to the drive will get written as soon as it is possible (or with a 
small amounts of data as soon as some timer reaches zero - this is to 
provide). But because your disk is able to write only let's say a 
20MiB/s and you are writing a 50MiB file the write without using cache 
would take about 2.5 seconds before your program can continue. So you 
are having a  25 times faster write at user level with cache than 
without. You could probably see it even better when reading lot of small 
files (ie. grepping a lot of text files), where the drive would have to 
seek for each request, but kernel can read ahead more data than was 
actually requested guessing that you will request them later anyway and 
put them in cache. And when you request them you get them at the speed 
of  your memory. So cache has (almost) nothing to do with swap and its 
on the same line as swap just because it fits better:P. Kernel is trying 
to be smart about swap and cache so it will swap out programs out of 
memory when you are really not using them to be able to use the memory 
for cache that can really speed up your system.

Well I believe this is correct, if not please somebody be so kind and 
correct me before beating me with a kernel tarball :).


Tomas Zvala



john moser wrote:

>Mem:    775616k total,   747740k used,    27876k free,    96584k buffers
>Swap:   250480k total,    71340k used,   179140k free,   298852k cached
>
>
>This is somewhat distressing.  Last time this happened, I started opening every
>program I had (including every OpenOffice.org product and every browser I had
>installed), and the "cached" value dropped quickly.
>
>I'm wondering, what IS cache?  It seems to increase even when swap is not used,
>and sometimes when there's no swap partition enabled.  It also seems to cause
>me to run into swap when I have ample ram available, assuming that cache is just
>some sort of cache that is copied from and mirrors another portion of ram for
>some sort of speed increase.  It's wasteful to me, and I want to more understand
>its implimentation and its purpose, and in all honesty limit its impact if possible.
>I got this RAM upgrade because I was using about 676M of RAM total, including swap,
>at peak; and now I find myself using 820M RAM at peak and about 750-800 continually.
>
>As always, i'm not on the list, so I'd prefer to be CC'd replies.
>
>_____________________________________________________________
>Linux.Net -->Open Source to everyone
>Powered by Linare Corporation
>http://www.linare.com/
>-
>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: Slab allocator . . . cache? WTF is it?

[Doug McNaught]

john moser <bluefoxicy@linux.net> writes:

> I'm wondering, what IS cache?  It seems to increase even when swap
> is not used, and sometimes when there's no swap partition enabled.
> It also seems to cause me to run into swap when I have ample ram
> available, assuming that cache is just some sort of cache that is
> copied from and mirrors another portion of ram for some sort of
> speed increase.  It's wasteful to me, and I want to more understand
> its implimentation and its purpose, and in all honesty limit its
> impact if possible.

Short answer: cache is the use of memory to store what would otherwise
have to be read from disk, like parts of frequently used files etc.
Linux will trade off cache against other RAM needs like program memory
on an ongoing basis, and (almost always) does a really good job of it.

Don't be alarmed about going into swap somewhat in normal
operation--Linux is storing less important data on the disk to make
more room in RAM for "hot" data.  Obviously, if you have hundreds of
megs in swap or it grows quickly, you may need to look at it.

> I got this RAM upgrade because I was using about 676M of RAM total,
> including swap, at peak; and now I find myself using 820M RAM at
> peak and about 750-800 continually.

That means the kernel is doing exactly what it's supposed to do.  If
memory isn't needed by programs and isn't used for cache, it just goes
to waste, so Linux tries to use it for one or the other.  :)

-Doug

RE: Slab allocator . . . cache? WTF is it?

[David Schwartz]

> Mem:    775616k total,   747740k used,    27876k free,    96584k buffers
> Swap:   250480k total,    71340k used,   179140k free,   298852k cached
>
>
> This is somewhat distressing.  Last time this happened, I started
> opening every
> program I had (including every OpenOffice.org product and every
> browser I had
> installed), and the "cached" value dropped quickly.

	So, as you've seen, when the kernel needs memory, it can easily throw its
cached data away and gain as much free memory as it needs.

> I'm wondering, what IS cache?  It seems to increase even when
> swap is not used,
> and sometimes when there's no swap partition enabled.  It also
> seems to cause
> me to run into swap when I have ample ram available,

	All true, all good. No matter how much RAM you have available, odds are you
are moving more dat to and from disk then you have RAM. So it sometimes
makes sense to swap stuff to make room for more cache.

> assuming
> that cache is just
> some sort of cache that is copied from and mirrors another
> portion of ram for
> some sort of speed increase.

	More typically, it contains copies of data that was written to or read from
disk in case it is requested (again).

> It's wasteful to me,

	Huh? Why do you say that?

> and I want to
> more understand
> its implimentation and its purpose, and in all honesty limit its
> impact if possible.
> I got this RAM upgrade because I was using about 676M of RAM
> total, including swap,
> at peak; and now I find myself using 820M RAM at peak and about
> 750-800 continually.

	If you don't want the computer to use the RAM, take the RAM out of the
computer. The computer will use all of the RAM you give it. Why shouldn't
it?

	DS

Re: Slab allocator . . . cache? WTF is it?

[Mike Fedyk]

On Tue, Dec 30, 2003 at 02:18:58PM -0800, john moser wrote:
> I'm wondering, what IS cache?  It seems to increase even when swap is not used,
> and sometimes when there's no swap partition enabled.  It also seems to cause
> me to run into swap when I have ample ram available, assuming that cache is just
> some sort of cache that is copied from and mirrors another portion of ram for
> some sort of speed increase.  

First of all, does it swap out and stay swapped out, or does it swap in and
out constantly?

What kernel version are you running?

The cache, or pagecache is at the core of Linux' memory management.  When
you start an application (or process) it first maps the file into the cache,
and then executes it.  So the cached value also counts the parts of the
files used recently.  It gets much more complicated, but you're invited to
learn more about it.  I'm sure someone else can give you a few URLs with
more details.

Re: Slab allocator . . . cache? WTF is it?

[Mike Fedyk]

On Tue, Dec 30, 2003 at 03:50:29PM -0800, Mike Fedyk wrote:
> First of all, does it swap out and stay swapped out, or does it swap in and
> out constantly?

http://www.matchmail.com/stats/lrrd/matchmail.com/fileserver.matchmail.com-swap.html

Look at the monthly graph.

Weeks 49 and 50 are 2.4.20-rmap.

Week 51 to half of week 52 is 2.4.23

After that it is 2.4.23-aa1.

I'd suggest that if you have any problems with too much swapping in and out,
to try 2.4.23-aa1.

Mike