IBM BS029ML - WebSphere Portal Server, Self Help Manual

Page: 46 / 242

32
IBM WebSphere Portal V6 Self Help Guide
2.5.3 WebSphere queuing mechanism
In order to understand how to maximize performance, it is necessary to understand the
WebSphere queuing mechanism. WebSphere implements a componentized architecture,
channeling requests through a number of queues. These queues or pools include a Proxy
server and Web server (considered even though they are external components), the
WebSphere Application Server embedded Web Container, the EJB™ container, data
sources, and possibly other connection pooling mechanisms to various custom back-end
systems. Each of these resources sustains a queue of requests waiting to use the resource in
question. The overall queuing mechanism is designed to converge towards the back end,
where resources are deemed more expensive. For example, out front it is not uncommon for
the Web server queue to be configured to handle an inordinately large number of requests.
This contrasts to a data source pool, which by nature is more expensive (both in terms of
CPU and memory) and thus usually only configured to handle a maximum of 10-20
connections simultaneously. Each queue has the potential to become saturated. There also
exists the possibility that if one of the back-end queues saturates, that it will have an effect on
the other queues in front. For example, it is not unusual that if a data source connection pool
saturates, that the Web container will also eventually overload (simply due to the fact that
requests cannot be processed further downstream). This can be particularly confusing when
investigating performance. In which case, it is recommended that you take a holistic approach
to performance tuning and determine which queue saturates first.
Figure 2-4 WebSphere queuing mechanism
Tip: Our experience has shown that many customers fail to implement vertical clustering
when horizontal clustering is implemented to address the needs of high availability. As
such, it is an IBM recommended best practice that both vertical and horizontal clustering
are implemented to address the needs of scalability, high availability, and operational
availability.
Web
Server
Web
Container
EJB
Container
Data
Source