Memory Buffers

Aaron Smith asmith at
Fri Nov 8 09:22:00 CST 2002

Oi.  The kernel VM Holy War rears it's ugly head again.  I figured
someone would tell me to upgrade the kernel.  Normally, I'm not adverse
to that.  I'm a little leery about doing it with this machine though. 
The reason being is that before it had the 2.4.14 kernel, it was running
a 2.4.9-ac10 kernel (compiled from the Redhat source rpm at the time). 
That kernel's VM caused no end of pain for us.  The machine would lock
up on a regular basis along with intermittently corrupting the database.
It was a nightmare.  The 2.4.14 kernel, however, contained a complete
re-write (or so I was told)of the VM and once that kernel was installed,
all the problems went away.  Both machines worked just fine for months. 
In fact, IIRC, that kernel was installed in April, and we didn't see the
buffer amount start to decrease until june/july, and the actual load
spikes didn't appear until September (once the average buffer size
decreased to the point were it could dip below 2MB.)
	From what I understand, the later kernels saw a move BACK to the VM
system that existed in 2.4.9.  Is that the case?   If so, I'll have none
of THAT thank you very much.  I've already been burned by it.  From what
I was able to gather from the Celebrity DeathMatch on the kernel
developer's list, Rik's VM was in 2.4.9, and Andrea's VM was in 2.4.14.
So what's in 2.4.17 that makes IT so darned special?  Is it just the
same code from 2.4.9 or is it perhaps a blending of the two to make
something better?  

On Fri, 2002-11-08 at 06:04, Arjan van de Ven wrote:
> On Thu, Nov 07, 2002 at 10:30:40AM -0500, Aaron Smith wrote:
> > 	Ok.  I have, what would seem to be, a fairly simple question.  When I
> > cat out the contents of /proc/meminfo on my Poweredge 4400, there is a
> > value for "Buffers".  What is this memory used for?  After doing some
> > research, all I've found is that they are I/O Buffers set aside by the
> > kernel.  My assumption, then, is that disk I/O is buffered into this
> > chunk of memory before being written out to disk.  Is this correct?
> > 	The REASON for the question is that our Poweredge 4400's (running
> > Redhat 7.1 and a non-rpm based 2.4.14 kernel) will suffer a massive
> 2.4.14 has a pretty bad vm. please use 2.4.17 or later, even if you want a
> self compiled kernel and not a RH kernel
Aaron Smith             		vox: 269.226.9550 ext.26
Network Director        		fax: 269.349.9076	
Nexcerpt, Inc.  
	...Nexcerpt... Extend Your Expertise

More information about the Linux-PowerEdge mailing list