Hi Sarma, What is puzzling me about the problem, is that if you have DEV with the same configured parameters, system statistics, and with a similar load (tables), the faulty code don't reproduce the problem on DEV. Apart from the bug cited, subtle plan variations (discarded as per you investigation), and other hidden things, honestly I can only now propose: - to "load" DEV with the most similar load as prod, and submit at "representative" number of connections. - thorougly check DEV kernel settings (compare the output of a "sysdef" on both HP-UX machines). Compare also the output of a glance or "top" for similar stack/resident memory usage for the processes involved in the "faulty" code. Check differences also at the HP-UX level between DEV and prod . regards.