Hi Mark, IMO it is easier to simply count the number of activations vs. the total number of accesses and infer the miss/hit ratios.
Regards, Djordje ________________________________________ From: Mark Meredith [[email protected]] Sent: Thursday, May 21, 2015 5:54 PM To: [email protected] Subject: Using Flexus to simulate DRAMSim rowbuffer hit rate? Hello, I am trying to use Flexus 4.1 to simulate DRAM (CMP.L2SharedNUCA.DRAMSim.OoO). I am interested in simulating rowbuffer hit/miss rate, so I added code in MemoryController::update() to log each DRAM command (precharge, activate, read, write, and refresh). I consider a "hit" to be two activate commands to the same rank, bank, and row, and a "miss" to be an activate command sent to a different row from the last activate command sent to the same rank and bank. Could someone let me know if this reasonable, to presume the behavior of the simulated rowbuffer from the sequence of commands processed in MemoryController.cpp? I'm not sure, the hit/miss rates from running Cloudsuite are not what I was expecting. Thank you, -Mark Meredith
