Hello all,
I am running a 2.4.0-test9 kernel and I have noticed a VM problem I
have not seen reported before. The machine is a uniprocessor Pentium
II with 2G of ram, and the kernel is compiled with CONFIG_HIGHMEM4G and
CONFIG_HIGHMEM both set to y. I also have 512M of swap on the machine.
I left a single large job running when I left yesterday afternoon
(size=1651M, RSS=1.5G). When I got in this morning I wanted to see if
it was still running so I typed "top" in an Xterm. When I hit return I
thought the machine had crashed. I could not move the cursor with the
mouse, or cause any other activity. I went and got a cup of coffe and
when I came back the machine was alive again. I then started netscape and
the machine appeared to crash again. It was completly unresponsive for
about 30 seconds and then it came back to life. As I type this message
into netscape, the machine will periodically freeze for about 30 seconds
and then come back to life. Anybody have any idea of whats going on?
On an unrelated note, is it possible for a process in 2.4 to see more
than 2G of address space? They seem to be limited to 2G for me. I was
hoping that the HIMEM stuff had removed that limit.
Thanks,
Jim
On Fri, Oct 27, 2000 at 07:03:29AM -0400, James Lewis Nance wrote:
> I left a single large job running when I left yesterday afternoon
> (size=1651M, RSS=1.5G). When I got in this morning I wanted to see if
> it was still running so I typed "top" in an Xterm. When I hit return I
> thought the machine had crashed. I could not move the cursor with the
> mouse, or cause any other activity.
I got a little more info. Even though the machine has 2G of ram and this
process'es RSS is only 1.5G, the rest of the memory is being used somewhere.
Top reports only about 1M as "free" memory. It also looks like kswapd is
running with a high CPU usage when this is going on. Its a little hard to
be sure since top freezes, but when it comes back to life kswapd shows up
near the top of the process list.
Thanks,
Jim
James Lewis Nance <[email protected]> writes:
> On an unrelated note, is it possible for a process in 2.4 to see more
> than 2G of address space? They seem to be limited to 2G for me. I
> was hoping that the HIMEM stuff had removed that limit.
You have 3GB user space address space. 1 GB is still reserved for the
kernel and you cannot break the 4GB limit for one process. But you can
have multiple processes using their own <3GB chunk of memory.
Greetings
Christoph