Why CPU and IO, not Interconnect, are necessary for RAC Performance

Here you see components that make up a typical RAC infrastructure with 2 nodes containing purely Oracle stack; not any third party components like Veritas Cluster. The nodes have been shown with the components that make up the stack in each. At the bottom is the Operating System that provides the operating foundation. Oracle Clusterware, aka Cluster Ready Services (CRS) sits on the O/S and brings the 2 machines into a cluster. CRS uses two very important devices – Oracle Cluster Repository (OCR) and Voting Device, which must be visible to both he nodes for the cluster to work. The third important component for CRS is the interconnect that binds the nodes together in a network. Interconnect is ideally on a separate switch. So, the cluster is a conglomeration of nodes tied together by interconnect network and the storage components such as voting and OCR devices.

ASM is the cluster-enabled volume manager for RAC, It uses CRS to provide the volume management to both the nodes, enabling the database to reside on the ASM storage. An ASM instance runs on each node. Then there is an Oracle instance that runs on each node. The caches of each instance exchange information on blocks (cache fusion), which goes over the interconnect as well. The global lock manager daemons running on each instance make sure the data resources are properly updated on the shared environment.

Service is a gateway to the RAC database. A database may have several services defined that control how the sessions connect to the instances and how they failover. Finally, Virtual IP (VIP) is a virtual IP address managed by the CRS that listens to the client requests. If a node is down, CRS moves the VIP to another node and sends out a no ack message when a client attempts to connect to it.

The red arrows show where the potential bottleneck points for the RAC system affecting performance. In the future slides we will explore the role these bottleneck points play and how to diagnose each.

Buffer Transfer Time

• The time is a sum of time for:
• Finding the block in the cache
• Identifying the master
• Get the block in the interconnect
• Transfer speed of the interconnect
• Latency of the interconnect
• Receive the block by the remote instance
• Create the consistent image for the user

Increasing Interconnect Speed

• Faster Hardware
  • Gigabit Ethernet; not Fast
  • Infiniband, even if IP over IB
• NIC settings
  • Duplex Mode
  • Highest Top Bit Rate (not Auto-negotiate)
• TCP Settings
  • Flow Control Settings
  • Network Interrupts for CPU
  • Socket Receive Buffer
• LAN Planning
  • Private LANs
  • Collision Domains

© 2011, www.oracledatabase12g.com. 版权所有.文章允许转载,但必须以链接方式注明源地址,否则追究法律责任.

相关文章 | Related posts:

1. RAC Performance Experts Reveal All
2. 11gR2 Grid Infrastructure Redundant Interconnect and ora.cluster_interconnect.haip
3. Configuring Linux for the Oracle 10g VIP or private interconnect using bonding driver
4. Oracle RAC Templates for Oracle Virtual Machine
5. Discover How To Leverage The Power Of The Grid
6. Adding a Node To 10gR2 RAC cluster
7. Demystifying Oracle RAC Internals
8. DW on RAC Best Practices