Sunday, June 29, 2008
Azul Systems and XML Processing
I was reading Peter Holditch blog on Use Cases for XML Processing and on one hand am happy that Azul was able to improve the performance of their customer's application, but on the other hand wonder why customers make stupid architectural mistakes...
Let's analyze Peter's comments:
OK, if this customer were smart enough to realize that there is a value proposition in going beyond commodity hardware to specialized Azul appliances that accelerate Java performance, then what prevented them from figuring out the same thing for XML?
I wonder if the better answer in this situation would have been to use IBM's DataPower appliance? Getting XML parsing and transformations out of software onto an appliance wouldn't just result in a 5x increase but in many cases, 100x increase! While I understand that Peter would have gotten fired by suggesting such an approach, it would be interesting to understand why the customer was asleep at the helm...
| | View blog reactionsLet's analyze Peter's comments:
- The system in question was a "SOA style" data movement application which extracts data from various systems, transforms it into a canonical XML format and from there passes it on to other downstream systems. A lot, but not all, of the data transfer is done in overnight batches.
At first blush, it was not clear how much moving to Azul might help this system. there was some parallelism, but not an overwhelming amount, and quite a lot of file i/o. This coupled with the fact that garbage collection during a batch process is not often a big problem (since there is no user present to complain if the system stops to garbage collect for a few minutes) made me wonder how much benefit Azul would show. Undaunted, we started to test.
OK, if this customer were smart enough to realize that there is a value proposition in going beyond commodity hardware to specialized Azul appliances that accelerate Java performance, then what prevented them from figuring out the same thing for XML?
I wonder if the better answer in this situation would have been to use IBM's DataPower appliance? Getting XML parsing and transformations out of software onto an appliance wouldn't just result in a 5x increase but in many cases, 100x increase! While I understand that Peter would have gotten fired by suggesting such an approach, it would be interesting to understand why the customer was asleep at the helm...
