IWA: Software offering that can be deployed on Linux/Intel
environment SMP, cluster, virtual and cloud environments. The host Informix database can be run on Linux/Intel,
HP-UX/Itanium, AIX/Power, Sparc/Solaris environments.
Exalytics: Runs on 4-socket Intel Westmere with 1TB (whether it runs Linux or Solaris) and can be combined to create a an exalytics pool. Would need use the specified hardware from Oracle to use this.
IWA: Designed to provide consistent performance without additional tuning and maintenance like index creation, statistics collection, cubing, pre-calculation of aggregates. Advanced algorithms achieve query on compressed data, exploit the SIMD instructions, on-chip cache, etc. See the details in this paper.
Exalytics: Performance is achieved In-Memory intelligent result set cache, In-Memory cubes and tuning the resultset caches, cubes, etc. Here is a quote from Oracle paper: For best query acceleration, Oracle Exalytics provides tools to analyze usage, identify and automate the pre-seeding of result caches. The pre-seeding ensures instant responsiveness for queries at run time.
Exalytics: Claims typical 5x compression. No information on whether the size comparison includes the size of the index/etc.
Exalytics: Needs Infiniband.
Supported Applications and compatibility.
IWA: Designed to accelerate application performance without needing any changes to application, reports or BI tools.
Exalytics: Designed to work with Oracle BI Foundation
only. Optional optimized version of Oracle BI
Foundation to improve the performance by 3X. Optional usually means additonal cost of acquisition. Essentially, if you're not a 100% oracle, Exa* platform shop, you can't use Exalytics. The interface seems to be via Essbase, because Oracle says Essbase applications will run as is. It also comes with a new data visualization tool.
Exalytics seems to be packgaing of Exadata (in the backend), TimesTen (for frontend and resultset caching), Essbase (for cubing services). There are additional technologies like in-memory columnar compression. The performance relies heavily on resulset caching (run the query once, save the results in-memory and when the same query comes again, return the query without actually executing), cubing. This means, the DBA will have to ensure right cubes are created, right results are cached, etc, etc. In-memory analytics should lead to less pre-calculation, not more. For tuning the workload, the 4-star wizard has to turn in to 5-star wizard to learn and tune all the new parameters, environment of Exalytics. Even with this, the best performance seems to be 15X to 20X.
for query tuning. Once you define the data that's accelerated, there's no tuning to do (like cubing, resultset caching).
The incredible performance comes from innovative architecture and patented algorithms. Because of this,
IWA can run in SMP, Cluster, virtual and Cloud environments providing this great performance.
Unlike Exalytics, IWA is designed to accelerate queries from custom applications and third party BI tools.