Non-Functional Requirements::High Availability: Product mappings - AIX

Product mappings correlate real products with one or more runtime nodes. In our high availability scenarios, we will document only the products implemented on nodes in the DMZ and external network, as this is where the high availability alternatives are implemented.

Recall also that the High Availability: Basic Runtime pattern is actually the Self-Service::Stand-Alone Single Channel: Basic Runtime pattern. The product mapping for this Basic Runtime pattern is not detailed below. As such, the first figure below corresponds to the High Availability: Runtime pattern: Single load balancer.

Variation 1: Single load balancer
Note that IBM WebSphere Edge Server V2 requires Java 1.3 to run correctly. Java 1.3 ships with WebSphere Edge Server for Windows and Linux. For AIX, the Java 1.3 runtime environment can be obtained at: http://www-106.ibm.com/developerworks/java/jdk/index.html

High Availability: Product mapping - AIX: Single load balancer
(Click a node to get a detailed explanation.)
Design Last Updated: 02-2004

IBM Edge Server
The load balancer node is implemented using the Load Balancer Dispatcher component of IBM WebSphere Edge Server V2.0. The Load Balancer component is configured so the two WebSphere Application Servers comprise one load balanced cluster. Requests can be distributed round-robin or based on load using advisors. The Media Access Control (MAC) forwarding method of Load Balancer can be used in this configuration However, both Network Address Translation (NAT) and Network Address Port Translation (NAPT) forwarding methods could be used. Enhanced recognition of available CPU and memory resources on content servers using advisors or metric server agents is provided with the Load Balancer. Session affinity to a specific Web server can be maintained using the sticky time affinity feature or the active cookie support of the Load Balancer. However, the use of application server clones using cookie based server affinity is also an option. For more information about maintaining session affinity, review the IBM White Paper WebSphere Application Server v4.x Best Practices using HTTP Sessions.

SSL connections from the client to the load balancer will be tunneled to the Web application servers.

WebSphere Application Server
The WebSphere Application Servers are configured with the same applications available on both servers. This can be accomplished using clones. Cookie based server affinity is provided by the WebSphere Application Servers.

Variation 2: Load balancer hot standby
There are two product mapping options available for the load balancer hot standby configuration.

Product mapping option 1
In this product mapping, two separate load balancer machines are used to perform the primary and standby load balancer functions.

High Availability: Product mapping - AIX: Load balancer hot standby: Option 1
(Click a node to get a detailed explanation.)
Design Last Updated: 02-2004

The Edge Server Load Balancer components keep in constant contact with each other by exchanging a heartbeat between them. This enables the standby load balancer to take over without losing IP packets.

Product mapping option 2
This product mapping option provides a load balancer hot standby configuration without increasing the need for physical machines.

High Availability: Product mapping - AIX: Load balancer hot standby: Option 2
(Click a node to get a detailed explanation.)
Design Last Updated: 02-2004

IBM Edge Server allows you to run a backup Load Balancer on one of the Web application server machines. This is referred to as collocation. Collocation of the WebSphere Edge Server Load Balancer component is supported on AIX, Linux and Solaris platforms.

In this configuration, the backup load balancer node could be affected by problems on the Web Application Server node. In addition, collocation of the backup load balancer function on a content server machine is not supported on all platforms.

Variation 3: Mutual high availability

High Availability: Product mapping - AIX: Mutual high availability
(Click a node to get a detailed explanation.)
Design Last Updated: 02-2004

Variation 4: Wide area load balancing
There are two product mapping options available for the wide area load balancing configuration.

Product mapping option 1

High Availability: Product mapping - AIX: Wide area load balancing: Option 1
(Click a node to get a detailed explanation.)
Design Last Updated: 02-2004

The Dispatcher component of the Load Balancer provides a Wide Area Network Dispatcher (WAND) enhancement that offers support for remote servers. A remote server consists of a remote load balancer machine and its locally attached back-end load balanced servers.

A client’s request packet can flow from the Internet to a load balancer machine, from there to a geographically remote load balancer machine, then to one of its locally attached back-end servers. The data returned to the client from the back-end server can flow directly back to the Internet and the client. This allows one load balancer cluster address to support all worldwide client requests while distributing the load to back-end servers around the world.The load balancer machine initially receiving the request can still have local servers attached to it and can distribute the load among its local back-end servers as well as the remote servers.

Product mapping option 2: Site selector
This variation is an alternative for wide area load balancing described previously. The remote clusters are load balanced. However, the load balancing is done by mapping DNS names to IP addresses, acting as a non-recursive (leaf node) name server.

High Availability: Product mapping - AIX: Wide area load balancing: Option 2
(Click a node to get a detailed explanation.)
Design Last Updated: 02-2004

Site Selector is a component of WebSphere Edge Server Load Balancer. It communicates with the Domain Name Server (DNS) to return the appropriate IP address to the client based on feedback from the back-end servers. It selects the "right" cluster or back-end server IP address to satisfy a client name service request. This allows the client to connect directly to the cluster or back-end server for all its communications.

Site Selector can be used in conjunction with advisors and the Metric Server component of Network Dispatcher to determine the best server that a client request should be sent to. Additionally, if Site Selector is used with Metric Server, it can make use of a feature called Network Proximity, that allows Site Selector to choose the best server according to the shortest response time of a ping between the servers and the client.

You should note the following considerations:

• DNS query responses might be cached at various places, delaying the change of Site Selector’s preferred target servers. Browsers will often cache DNS resolved IP addresses.
• Site Selector depends on DNS servers for which it acts as a leaf node.

Variation 5: Caching proxies with security plug-in
The following product mapping figures include all the software that is required on each of the primary nodes to implement this high availability caching proxy Runtime Pattern.

High Availability: Product mapping: AIX - Caching proxies with security plug-in
(Click a node to get a detailed explanation.)
Design Last Updated: 02-2004

The load balancer nodes are implemented using the high availability Load Balancer feature of WebSphere Application Server V5 Edge Components. The Load Balancer feature is configured so the two caching proxy nodes comprise one load balanced cluster. Requests can be distributed round-robin or based on load using advisors. The Media Access Control (MAC) forwarding method of Load Balancer can be used in this configuration. However, both Network Address Translation (NAT) and Network Address Port Translation (NAPT) forwarding methods could be used. Enhanced recognition of available CPU and memory resources on content servers using advisors or metric server agents is provided with the Load Balancer. Session affinity to a specific back-end server can be maintained using the sticky time affinity feature or the active cookie support of the Load Balancer. SSL connections from the client to the load balancer will be tunneled to the back-end proxy servers. For additional information on performance, scalability, and high availability within WebSphere Application Server, see IBM WebSphere V5.0 Performance, Scalability, and High Availability, SG24-6198.

Next, Review guidelines and related links or review another product mapping: Windows 2000