G
George Potemkin
Guest
The customer run on HP Superdome. Its architecture makes the access to the shared memory slower. There are no flushes of DB blocks. I'm not sure if Progress flushes BI and AI buffers. Anyway there are no much to flush: -bibufs/-aibufs 20 and the number of the full BI buffers is about 2 with max value equals to 5. At checkpoint Progress creates and writes the RL_INMEM note with the compressed list of the opened transactions. In this case the list is rather large: 200-500 transactions but I think it still requires just a small part of second to write the RL_INMEM note.
Continue reading...
Continue reading...