On Wed, 10 Mar 2010 14:30:58 -0600, "Boehm, Matthew" <mboehm@stripped> wrote:
> I'm looking for comments on monitoring swap usage on ndbd nodes. I've
> got two NDBD nodes that have 16GB of RAM. 10GB are dedicated to
> DataMemory and 2GB to IndexMemory. I've got lockpagesinmainmemory set.
> That leaves about 4GB left to the OS of RAM plus another 4GB of swap.
> In our managed hosting, we monitor swap usage and send alarms on certain
> usage thresholds. For this one ndbd node, all the swap gets used up at
> some point but still plenty of RAM left and system perfectly stable.
> Should we even then bother with monitoring the swap usage? Turn it off
> completely for ndbd nodes? Lower the vm.swappiness?
In the past, I've seen systems get rather swap happy (at least paging
things out) when large amounts of memory is locked (such that NDB does).
So it can depend on what kernel you're running... as there has been
But tweaking the VM parameters is one possibility - basically try and
make it behave as if it had that 4GB of memory, not as if it has 16GB.