The problem with large Java heaps and long-lived objects (such as cached objects), is GC pauses can stop the world for long periods of time. In fact, this is what made us look at memcached in the first place -- to replace our simple HashMap-based cache. Java's GC is always improving, but memcached works in O(1) time today. Not to mention memcached's nice distributed cache possibilities.
-chris On Thu, Apr 30, 2009 at 1:19 AM, Mohan Radhakrishnan < [email protected]> wrote: > > Hi, > I have a basic question. We are planning to buy expensive > hardware like HP-UX or IBM-AIX or Windows. The processors are 64-bit. > I checked the JVM heap limitations and found that upto 4GB is > supported. > > I also know that memcached cache farms can scale better than the > Java heap. What is the deciding factor here ? Can't the java heap( 2 > cluster members ) itself be used as a cache ? I know that in a web > application we are limited to some type of scope like the session for > caching. > > How would I go about evaluating our requirements to support a large > number of users( 300000) ? > > > Thanks, > Mohan >
