what is the difference between DPF and PureScale?

In short: many instances each accessing its own fraction of data versus many instances accessing the same data.

As mentioned by n_i, DPF is based on a share nothing architecture, which is more appropriate for data warehouses (which is why it is now called InfoSphere Warehouse). Share nothing architectures like DPF are used in parallel database applications, where each query runs in parallel on each node (accessing the data for a particular table that each node contains) and the coordinator node "glues" the answers from each node back into a single answer to be return to the SQL statement requesting the query.

PureScale is based on a shared disk architecture, more appropriate for OLTP.

so if we have mixed environment (datawarehouse/OLTP) which one would be preffered?

MarkhamDBA, Not sure if this will help or not but here is a linke from V9.8 infocenter:

Comparison of a DB2 pureScale environment with a multi-partition database environment

For what purpose?

If you have OLTP and data warehouse queries hitting the same tables, DPF is not appropriate (too much overhead for OLTP). If they were different tables, you could partition the warehouse tables with DPF and not partition the OLTP tables in the same database.

PureScale provides hardware scalability and flexibility (ability to grow your server(s) as needed beyond just one server for a given OLTP database) and some degree of fault tolerance (since you can loose one server and still keep going), but does not provide fault tolerance for storage since it uses shared disk (unlike HADR). PureScale does not provide any serious performance advantages (other than the ability to scale the hardware as needed) if your computing needs are fairly constant or the hardware can be scaled up without adding a whole new server.