xxxxx@xxxxx.com 2001-11-16
The fix to
Bug Id: 4314534 Synopsis: NFS cannot be controlled by SRM
has resulted in a call to thread_create from user land
for each nfsd threads.
The call to thread_create uses default stacksize which
means that 1MB will be used. Since nfsd is a 32-bit process
it means that the number of active nfsd thread would be limited
by the available address space. Since the user level stack is
mostly unused, its would be easy to specify a small value for it.
There is a more important aspect to this though.
For each user thread, there will be a kernel lwp created
with default stack size: lwp_default_stksize.
This means that rpcmod:svc_default_stksize becomes ineffective
in controlling the kernel side stack associated with nfsd threads.
This could have serious implication to running NFS over VxFs and
possibly to xxxxx Cluster installation, PxFs and Tivoli. (see associated
bugid). Those product recommend tuning rpcmod:svc_default_stksize and this
now have no effect on nfsd threads.
xxxxx@xxxxx.com 2001-11-16