Enhancing performance of BLOB loading

Question

Hi,
I am loading one row of BLOB data using DBMS_LOB Package. To load 5 GB of BLOB data its taking 8:53 Minutes. Due to some restriction in the target database, SQL *Loader can't be used.
Is there any other option/method to fasten the performance of the above said BLOB load? Any option with you is welcome except SQL *Loader. Thank you.. :)                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       

Answer 1

user637568 wrote:
Is there any other option/method to fasten the performance of the above said BLOB load? Any option with you is welcome except SQL *Loader. Thank you.. :)Does not matter what you use, there are a bunch of static layers here - and these impacts the performance mostly.
On the client side - the data of the LOB needs to be read from where? Disk? Mapped drive? The speed the client can read this data is largely dependent on how fast that storage can deliver the data to it.
The client needs to ship that data across to the Oracle server process. What is the network latency between client and server? How many routes and hops do it take for a client packet to reach the server?
The client's Oracle driver packages LOB payload into TCP packages. How effective is this? Does it stuff a TCP packet as full as possible? Ideally for network performance, one wants large packets carrying the LOB data to the server - and not a gazillion small packets each with only a couple of bytes of LOB data.
On the server side, Oracle needs to write that LOB data stream it receives to disk and commit it. Just how effective is I/O on the storage used by Oracle? How many other processes are hitting that same I/O layer?
You need to look at all these layers and ensure that each one of these is configured and used optimally. The elapsed time of the end-to-end process will be determined by the slowest layer. In such a case, one can consider using parallel processing (assuming that the layers have the capacity).
For example, the client reads the source data, sees it is 8GB in size and decides to use 16 processes to each read and transfer 512MB concurrently to the server - where the PL/SQL code on the server side has the intelligence to re-assemble these 16 chunks (in 16 different client sessions) into a single 8GB chunk for final storage.