https://issues.apache.org/bugzilla/show_bug.cgi?id=56798
--- Comment #5 from Christopher Schultz <ch...@christopherschultz.net> --- (In reply to Sebb from comment #3) > (In reply to Christopher Schultz from comment #2) > > > Using a > > LIFO structure does in fact mean that the whole pool will stay "hot" > > regardless of how big it becomes... as long as the connections are used > > frequently enough. > > Surely you mean FIFO here? Yes, I meant FIFO. Sorry for the confusing typo. > > But the frequency of a check-out is not really relevant: only the number of > > connections that are used /simultaneously/ is relevant. A site may be able > > to get away with a pool with a depth of 2 or 3 connections, but if the pool > > grows to 50 connections, they all might stay in the pool and not be evicted > > due to the LIFO behavior. > > Ditto > > > I'm not sure if the LIFO queue gives a performance benefit -- because the > > head of the structure is only under contention for check-outs and the tail > > of the structure is only under contention for check-ins > > Huh? Is that really true of a LIFO queue? A queue is FIFO, a stack LIFO. To align the nomenclature, here, queue is FIFO / LRU and stack is LIFO / MRU. For a stack/LIFO/MRU structure, there is only one "hot" end of the structure: both check-ins and check-outs are happening in the same place (the "top" of the stack). For a queue/FIFO/LRU structure, check-ins are happening at one end and check-outs are happening at the other end. So, depending upon the synchronization techniques used, the queue may offer better performance at the structure level -- even if the stack model gives a better outcome. It was just a thought. -- You are receiving this mail because: You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
