Customer White Paper on Active GridLink
By Steve Felts-Oracle on Aug 11, 2013
There's a new white paper available at http://www.oracle.com/technetwork/middleware/weblogic/overview/activegridlinkwhitepaperoraclenec-1987937.pdf that is jointly written by NEC and Oracle.
It starts out by covering terms and features in Oracle RAC, Oracle database 11gR2 and 12c, and WebLogic Server Active GridLink (AGL). The remainder of the paper focuses on some detailed test scenarios of the 11gR2 Oracle database features with Active GridLink. These are real-world testing results (not all of them turn out as expected) on some fast hardware. The results include some interesting graphs and tables. I will include some summary remarks here but I recommend reading the full paper.
You can use a Single Client Access Name (SCAN) address to take advantage of adding a RAC node without changing the WLS configuration. SCAN combined with Runtime Connection Load Balancing (RCLB) evens out the connection load across the currently available instances using load balancing advisories.
Fast Connection Failover (FCF), which you get by default with AGL when Fast Application Notification (FAN) is enabled, can significantly improve response to abnormal events like loss of the public network (15 second response) or loss of the interconnect (33 seconds) compared to waiting for a TCP keep-alive timeout. However, it needs some network to be available on which to receive event information. It also didn't help when a process stalls. Other Oracle timeout mechanisms can be used to help free up the client in these situations (e.g., setting oracle.jdbc.ReadTimeout). There is a reference in there to NEC ClusterPro to solve the stalled process case - the reference is in Japanese but an internet search will turn up some English documentation.
There are some interesting tests with setting RLB_GOAL to SERVICE_TIME or THROUGHPUT to enable RCLB. When additional load is added to one of the instances, the response time initially goes up significantly but RCLB evens it out over time by reducing the connections used on the overloaded instance. However, an issue was found in the case that the load is high (over 80%) that the performance does not improve as expected when using SERVICE_TIME.; this seems to lean toward using THROUGHPUT for the RLB_GOAL setting (this issue is still being investigated).
Web Session affinity can significantly reduce interconnect traffic between the RAC instances and thus reduce the related response time (in one test, the response time was cut in half and interconnect traffic basically disappeared). Also note that session affinity works across session replication in a cluster so even if there is a WebLogic Server failure, cache fusion is reduced when continued on another server with the same session affinity.
The conclusion is that AGL easukt provides a lot of complex functionality. Features like session affinity and RCLB is complete transparent to the application (no coding). AGL fits in well with existing applications. If you look at this earlier blog Migrating from Multi Data Source to Active GridLink, you will see how easy it is to update the configuration to move from Multi Data Source to AGL and no changes are needed to the application. AGL provides high performance, availability, scalability and serviceability,
If you are running WebLogic Server 10.3.4 or later and using RAC 11gR2 or later, check out Active GridLink..