4-core slow due to disk access

[Rashkae]

Joep L. Blom wrote:

> My outputs are:
> ~$ free
>               total       used       free     shared    buffers     cached
> Mem:       3992748    2895328    1097420          0      50168     847436
> -/+ buffers/cache:    1997724    1995024
> Swap:      8193108     889160    7303948
> 
> and:

> Committed_AS:    3019596 kB
> VmallocTotal:   34359738367 kB


That looks like a comfortable load.  Though I'm impressed at anyone 
other than me who an drive a desktop to 3GB Allocated... :)

The amount of swap used still surprises me, but that might be due to 
simple kernel tuning on the kernel version you are using.

My next suggestion would be to use some kind of swap monitor and watch 
it while this disk activity is slowing you down, and see if you can 
confirm if the system is hitting swap at those times.  My personal 
favorite is Gkrellm. It has nice visual charts that will often show you 
at a glance what your bottleneck is.  For something more old fashioned, 
you can use vmstat (part of the sysstat package.)..  The usual sign that 
you are dealing with memory starvation is when the system has to swap 
out and swap in (read / write to swap) at the same time. I believe that 
to be an indication that you ran out of ram for the task at hand and 
memory needs to be evicted from ram to make room for the pages being 
requested.  As a strictly rule of thumb, which is by no means scientific 
and I'm sure would be greatly debated, I don't usually observe that 
problem until Committed_AS: value exceeds physical RAM.  If that is the 
case, however, then it becomes a matter of finding out which process 
exactly chewed up all that memory.

I see from your free that you have 800MB of swap used even though there 
is still 1 GB of ram free,  I suspect that a process was earlier taking 
up all that memory which was then terminated (or crashed)

(PS, on most system, the Committed_AS: on a freshly started computer and 
gnome desktop should be in the neighbourhood of 500MB)