RAC on HP-UX Highlighted Recommendations

RAC on HP-UX Highlighted Recommendations

This is a list of the core best practices and references identified by the RAC Assurance Team relevant to HP-UX specific RAC implementations. The rest of this chapter provides additional best practices and references from our team.

• Hostnames on HP-UX should be 8 characters or less. Refer to Note 469569.1 for more details.
• Setting streams_sqmax=50000 causes gc cr block lost to be reduced. The default value of the 'streams_sqmax' was 'unlimited' for HP 11.11 or lower version, for HP 11.23 the default was changed to 1000.
• On HPUX (PA-RISC) and HPUX (Itanium) and RDBMS Versions 11.1.0.6 when the DIAG process is activated to get the call stack for a background process on the instance, under certain situations (corrupt call stack on the target) this can result in the target process for diag being killed. If the target process is a critical background process (such as LMS) this can also result in the instance being terminated.
  

Workaround: Set the hidden parameter "_ksb_disable_diagpid" = true to prevent diag from taking stack traces from other processes. Note: This may result in insufficient diagnostics to progress a problem should there me an issue needing such information. This is fixed in 11.1.0.7. Reference: Note 6489596.8 and Note:6849769.8

• Following is for HP-UX (Itanium) only: Implement the workaround mentioned in knowledge document Note 763724.1 RACGMAIN received SIGSEGV on checkresource causing a crash of a resource.
• For versions 10gR2 and 11gR1, it is a best practice on all platforms to set the CSS diagwait parameter to 13 in order to provide time for dumping diagnostics in case of node evictions. Setting the diagwait above 13 is NOT recommended without explicit instruction from Support. This setting is no longer required in Oracle Clusterware 11g Release 2. Reference Document 559365.1 for more details on diagwait.

=========================
11.2.0.2
=========================

• NTPQ executable should be in the grid install user's path, if ntp is being used.
  

RAC on HP-UX Best Practices

Operating System 

• In order to assure proper Oracle process scheduling priority for HP-UX, review Note:759082.1
• In HP-UX Itanium, RACGMAIN can cause a crash with a SIGSEGV error - Note:763724.1.  This can occur in 10.2.0.4 and 11.1.0.7. 
  

Storage

• Cannot discover Disks in ASM after upgrade to 10.2.0.3 on HP-UX Itanium – Note 433770.1. Apply patch PHCO_35524 or rename /usr/sbin/lvmchk to something else or apply oracle patch 6051728 or upgrade to 10.2.0.4.
• Shutdown the CRS stack cleanly before trying to unmount the shared filesystem. Most other clustered file systems require you to stop all process holding file descriptors open, before permitting the unmount command to continue. That is not presently the case with HP-UX Veritas Clustered Filesystem. If the clustered filesystem with OCR is unmounted 'cfsumount,' whilst the crs stack is still up and running, the cfsumount kills any holding processes. This terminate the OCSSD process, without warning

Networking

• Jumbo frames with mtu 9000 cause problems with hp nc7170 drivers.
• Setting streams_sqmax=50000 causes gc cr block lost to be reduced.  The default value of the 'streams_sqmax' was 'unlimited' for HP 11.11 or lower version, for HP 11.23 the default was changed to 1000.
• Note: 883801.1 - Ensure Patch PHKL_40208 applied to avoid RAC hangs on HP UX Itanium 

Configuration

• On HP-UX Itanium platform and bash version 3.2.0(1), the following problem occurs:
  sbin/init.d/init.cssd[995]: 21571 Memory fault
  /sbin/init.d/init.cssd[1001]: 21579 Memory fault
  Workaround:  Use /usr/bin/ksh or bash version 3.0.   Reference Note 734618.1.
  

Install

• Installing RDBMS Software on Itanium Servers with Montecito Processors – Note 400227.1 and Note 404248.1

Patching

• Make sure patch phne_35766 s700_800 or its superceded patch 11.23 cumulative arpa transport patch is applied. 
  This patch is reccomended because it can result in network packets being dropped intermittently or continuously resulting in the failure of network applications or severe network disruption. 
  http://www13.itrc.hp.com/service/patch/patchDetail.do?patchid=PHNE_35766&admit=109447627+1242314169748+28353475
• During migration from Oracle 10.1 to 10.2, installer does not copy the vendor clusterware library to all nodes of the cluster.  Note 392007.1 provides the workaround.