Tune > Performance
Improve performance with WebSphere eXtreme Scale V7.1
IBM WebSphere eXtreme Scale v7.1 is a distributed caching solution based on an in-memory grid that manages data and business logic across servers.
In a properly tuned WebSphere Commerce environment, most page view requests are served directly from memory cache rather than from slower physical disk.
To ensure low response times and high transaction throughput, large amounts of server memory are allocated to application cache, within which are stored Commerce artifacts such as...
- page fragments
- content objects
- JEE presentation layer elements
In a clustered environment where there are multiple instances of Commerce running, each with its own memory space dedicated to cache for that particular cluster member, cache capacity is limited by the individual JVM size within the cluster. In effect, the total size of any given cache is statically limited to the maximum size as defined in the application server instance.
As nodes are added in order to scale application capacity, each cache associated with a JVM becomes a replica of all the other instances. A side effect of this design is that as each new cluster member starts up, it must build its own cache replica from scratch, which places significant stress on the shared database, and remains limited in size regardless of the numbers of nodes added.
WXS addresses a number of design contraints...
- Elimination of disk storage as a performance bottleneck.
One of the potential performance constraints in a Dynacache environment results from processing overhead incurred during the swapping of cache objects paged to disk. Since cache sizes are greatly increased, fewer disk operations are required and potential saturation of available CPU is avoided, resulting in higher throughput and lower system utilization.
- Reduction of processing overhead and latency associated with exchange of invalidation related information between physically separate cache instances.
As WXS instances are managed as a single addressable memory space without physical partitions between instances, the following benefits accrue...
This results in...
- Reduction of network bandwidth penalties associated with checking and invalidating cache objects
- Reduction in CPU usage
- Allows changes to product information in real time
- Significant reduction in database access when new cache instances start up by elimination of cold start database access spikes.
In a dynamic clustered environment where nodes are brought up and down on a regular basis, WXS dramatically not only reduces the stress placed on the database tier when new cluster members come on line, but WXS will "warm" the new server's cache memory in a fraction of the time by pre-loading all new cache instances from an existing "hot" set of data that is a subset of the in-memory records, and lazily load the remaining data.