SAFARI Technical Report No. 2015-004 (March, 2015) The Blacklisting Memory Scheduler: Balancing Performance, Fairness and Complexity Lavanya Subramanian, Donghyuk Lee, Vivek Seshadri, Harsha Rastogi, Onur Mutlu Carnegie Mellon University flsubrama,donghyu1,visesh,harshar,
[email protected] SAFARI Technical Report No. 2015-004 Abstract—In a multicore system, applications running on different cores is application-aware memory request scheduling (e.g., [30, interfere at main memory. This inter-application interference degrades 31, 28, 17, 18, 41]). The basic idea of application-aware overall system performance and unfairly slows down applications. Prior works have developed application-aware memory request schedulers to memory scheduling is to prioritize requests of different ap- tackle this problem. State-of-the-art application-aware memory request plications differently, based on the applications’ memory ac- schedulers prioritize memory requests of applications that are vulner- cess characteristics. State-of-the-art application-aware memory able to interference, by ranking individual applications based on their memory access characteristics and enforcing a total rank order. schedulers typically i) monitor applications’ memory access In this paper, we observe that state-of-the-art application-aware characteristics, ii) rank applications individually based on memory schedulers have two major shortcomings. First, such sched- these characteristics such that applications that are vulnerable ulers trade off hardware complexity in order to achieve high performance or fairness, since ranking applications individually with a total order to interference are ranked higher and iii) prioritize requests based on memory access characteristics leads to high hardware cost based on the computed ranking. and complexity. Such complexity could prevent the scheduler from We observe that there are two major problems with past meeting the stringent timing requirements of state-of-the-art DDR pro- tocols.