Trace Kernel Memory Consumption

[Todd Urie]

I am trying to resolve a problem where my NFS server experiences sudden
increases in kernel memory usage. For example, over the course of a single
5 min polling interval the kernel consumption goes from ~35% to > 80%. On 3
of the 4 occasions when this has occurred the server has completely
exhausted it's memory to the point where it becomes unresponsive and can
only be recovered via break and reboot.

I know enough about Dtrace to get myself in serious trouble but that is
about it. Does anyone know of any methods that I can use to determine what
is allocating kernel memory and how much?

[David Blasingame Oracle]

You might want to check this thread

http://opensolaris.org/jive/thread.jspa?threadID=87240

You can turn on kernel memory auditing with kmem_flags as described in
the thread or use the fbt provider in which kernel memory allocations
occur kmem_alloc(). Both described in the thread.

Dave

Post by Todd Urie
I am trying to resolve a problem where my NFS server experiences
sudden increases in kernel memory usage. For example, over the course
of a single 5 min polling interval the kernel consumption goes from
~35% to > 80%. On 3 of the 4 occasions when this has occurred the
server has completely exhausted it's memory to the point where it
becomes unresponsive and can only be recovered via break and reboot.
I know enough about Dtrace to get myself in serious trouble but that
is about it. Does anyone know of any methods that I can use to
determine what is allocating kernel memory and how much?
--
-----RTU
------------------------------------------------------------------------
_______________________________________________
dtrace-discuss mailing list