With OLTP applications scalability and performance are largely due to storage-related bottlenecks. MIT analyzed OLTP performance and found that very little of the system resources were consumed by actually executing business-process-level transactions. Nearly half of the CPU cycles were consumed managing around under-performing storage by either managing a complex buffer pool or dealing with storage related recovery issues.
Working around OLTP application problems is something that has become status quo but at a great cost. But the problems and challenges are growing beyond the continued growth of OLTP databases.
There are more implementations of OLTP applications on virtual servers, which creates an even greater I/O bottleneck due to sharing physical server infrastructure with multiple applications.
Data warehouses and business intelligence environments are moving to near-real time refresh cycles, which places additional extraction burden on OLTP applications.
When near-real-time isn't good enough, parts of the organization are demanding the ability to run "operational reporting" - to execute queries on live transactional applications during business hours.
Next generation process-oriented OLTP applications are being built that want to wrap them into various kinds of services oriented models that increase both read and write traffic against the DBMS. This creates an even bigger storage performance bottleneck.
