Concerns and issues relating to all versions of WebSphere Application Server
I am excited about the new Dynacache heap management feature in WAS7. The primary motivation for introducing this feature is for serviceability. Many times customers trip up on themselves by caching too much stuff. I have lost count of the number of critical situations where I have been hauled from the bed, by some customer critical situation where the server was down due to an OOM error. It is extremely important to size the cache well. Unfortunately this is not an easy thing to do and pre-production stress testing in some cases does not simulate real world traffic which leads to an unutilized cache and thus false confidence in the capacity utilization of the cache.
This is not an easy problem to solve. There is no sizeof operator in Java which will tell us the size of an object in the JVM heap. Therefore we have to use all sorts of smarts, trickery and some help from the application developer to determine the total amount of memory on the JVM heap taken up by the cache. All earlier techniques to determine cache size rely on serializing the cached objects and metadata because that is the only way to accurately determine the size of the objects. In WebSphere 7 we have taken a much light weight patented approach which does not rely on serialization to determine cache heap size.
Most application servers allow cache size to be controlled by no.of entries. We are taking cache size management to the next level in WebSphere Application Server 7.
What exactly does Dynacache provide ?
WAS Dynacache component will provide an ability to constrain the cache in terms of the JVM heap. In addition to specifying the cache size in MB, Dynacache will also allow customers to set a high water mark and a low water mark for the cache heap consumed. Once cache heap memory reaches the high water mark, dynacache will either discard or evict the least recently used items to disk, till the cache is brought down to the low water mark. This functionality of limiting the cache in terms of the JVM heap will be available if the objects put into the cache that implement the com.ibm.websphere.cache.Sizeable interface. When servlet caching is enabled, all the cached JSP and servlet responses will be Sizeable. This interface will have one method which will return the size of the Object in bytes put into the cache. Dynacache will use the Sizeable interface to estimate the heap size of the cache. This feature will be OFF by default. A customer will have to explicitly enable and set cache limits and watermarks.
How to enable this feature ?
On the Dynacache service panel, WebSphere exposes the Dynamic Cache object cache service and Dynamic Cache servlet cache service. The Dynamic object cache service is always started at server start up. The Dynamic servlet cache service is started when the servlet caching is enabled in WebContainer panel. There is now a checkbox for Memory Cache Size which will control the memory cache size with high/low threshold. In addition to specifying the size the customer can also specify a range for the heap by setting threshold limits. The Servlet and Object Cache instance under the Resources -> Cache Instances have also include the Memory Cache Size control feature.