Besides that, partitions allow you to dynamically change your interest: depending on the situation, your application can decide to leave one partition and attach to anoter one. Or it could decide to just add another partition to the list of partitions it is collecting its data from. If every partition would have a separate topic, the application would need to manage a variable number of topics, each one having its own separate reader and writer. As you can see, just having a list of logical partitions to which your publisher and subscriber are attached is much more flexible.
posted on Thursday, May 06, 2010 - 02:18 pm
can we talk about hardware partitionning (RAM, CPU...)?