Database Reference
In-Depth Information
Highly Available VIP (HAVIP)
Version 12c introduces the new feature HAVIP, a Clusterware-monitored VIP that can be used for non-Database
applications. Typically, VIPs are created for VIP or SCAN listeners for Database connections. In some cases, the need
arises to create a VIP monitored by Clusterware designed to support other applications.
For example, following command creates a new VIP resource in the Clusterware. Starting this VIP resource will
plumb a new IP address in a node. The following command adds a VIP resource with an IP address of 10.5.12.110 in
the Clusterware. Enabling the resource plumbs IP address 10.5.12.110 in rac2.example.com node. Note that this IP
address is added to the second network, indicating that HAVIP can be added in any network.
# srvctl add havip -id customapp -address 10.5.12.110 -netnum 2
# srvctl enable havip -id customapp -node rac2.example.com
This HAVIP resource was primarily designed to support highly available NFS export file systems. The following
command adds an NFS-exported file system using the HAVIP resource created earlier.
# srvctl add exportfs -name customfs1 -id customapp -path /opt/app/export
You could also create this file system automatically exported to clients, as shown in the following command.
In this case, NFS is exported to hr1 and hr2 servers. 10
# srvctl add exportfs -name customfs1 -id customapp -path /opt/app/export -clients hr1,hr2
The HAVIP feature is a new feature to extend Clusterware functionality, to implement and maintain highly
available non-database VIP and export file system.
Kernel Parameters
Kernel parameters affecting network configuration are a critical area to review for Oracle RAC. Misconfiguration or
undersized parameters can cause noticeable performance issues, even errors.
In the Linux platform, rmem_max and rmem_default control the size of receive socket buffers, and wmem_max and
wmem_default control the size of send socket buffers. If the interconnect usage is higher and if you have plenty of
memory available in the server, then set rmem_default and wmem_default to 4M, and set rmem_max and wmem_max
parameters to 8M. However, these four parameters control the size of socket buffers for all protocols. In the Solaris platform,
udp_xmit_hiwat and udp_recv_hiwat parameters control the size of socket buffers for UDP, and the documentation
recommends setting these two parameters to at least 64K. Our recommendation is to increase these parameters to
about 1M.
Memory allocated for fragmentation and reassembly should be configured optimally too. If the buffer size is
too small, then the fragmentation and reassembly can lead to failures that lead in turn to dropped packets. In Linux,
parameters net.ipv4.ipfrag_low_thresh is the lower bound and net.ipv4.ipfrag_high_thresh is the upper bound
for memory allocated for fragmentation and reassembly.
The parameter net.ipv4.ipfrag_time determines the duration to keep IP packets in the memory waiting for
reassembly, and the value of this parameter defaults to 60 seconds. If the IP packets are not assembled within 60
seconds, for example, a required IP packet is not received, then all IP packets of that UDP segment are dropped.
However, this timeout is probably not effective in Oracle RAC, since database-level timeout is configured to be smaller
than 60 seconds. We will explore database-specific timeout while discussing go lost packets wait event.
10 Note that this is not a complete set of steps to create NFS exported file system.
 
Search WWH ::




Custom Search