HP Tru64 UNIX and TruCluster Server Version 5.1B-5 Patch Summary and Release Notes (March 2009)

• Corrects a scenario in which a spurious wakeup is sent to a process that had
interrupted a blocked attempt to set a mandatory file lock.
• Provides performance enhancements to the vm_overflow feature.
• Changes the way page migrations occur on a NUMA system to address poor
performance due to excessive paging.
• Corrects an incompatibility between the cpus_in_rad and
gh_chunks/rad_gh_regions tunable attributes that could result in a boot failure.
• Corrects a problem with pagetable page allocations that could leave a thread
waiting indefinitely during a fork operation.
• Corrects "ubc_wire: hash failed" panics on non-NUMA systems.
• Corrects a "not wired" panic that occurs with System V shared memory and
bigpages.
• Reduces scheduling contention when unmapping shared address space that has
an extremely high number of mappers.
• Handles reservation conflict errors to address cluster node hang during boot.
• Updates the audit system to display additional information for numa_syscalls and
msfs_syscall system calls.
• Provides a tunable attribute to allow the reserving of a percentage of vnodes for
root use.
• Fixes an issue in the VM subsystem in which a page that is not managed by VM
is incorrectly identified as being managed by VM.
• Fixes a problem in UFS superblock update logic that could result in a failure to
report errors encountered when writing cylinder group summary blocks.
• Provides a new cluster-specific link aggregation distribution algorithm when using
LAG in a LAN cluster.
• Reduces fork/exec overhead on systems with large default stack sizes.
• Fixes a cluster deadlock/hang issue when a new device is discovered.
• Fixes a kernel memory fault panic that occurs during a NetRAIN interface failure.
• Fixes a CPU hang caused by /dev/random.
• Corrects a condition that causes a process hang and system panic when using
System V shared memory and asynchronous I/O.
• Corrects a minor tuning issue with netisrthreads on NUMA machines. Previously,
four netisr threads were started per RAD, whereas the default number of netisr
threads per RAD now equals the number of CPUs per RAD.
• Fixes problems with RPC for configurations with no RAD 0.
• Fixes a problem in which unkillable processes occur when debugging
multi-threaded programs with the totalview debugger on a Sierra Cluster.
• Fixes an underlying problem in NFS that can trigger an assert failure in CFS. These
conditions can be triggered by a failed NFS mount.
• Resolves a synchronization issue between pageout and exit paths.
3.4 Summary of Base Operating System Patches 119