dCCPI-predictor: A state-aware approach for effectively predicting cross-core performance interference
dCCPI-predictor: A state-aware approach for effectively predicting cross-core performance interference
复制标题
dCCPI-predictor：一种有效预测跨核性能干扰的状态感知方法
DOI：
10.1016/j.future.2019.11.016
复制
发表时间：
2020-04
影响因子：
7.5
通讯作者：
Robertas Damasevicius
中科院分区：
文献类型：
--
作者：
Jingwei Li;Yong Qi;Wei;Jinwei Lin;Marcin Wozniak;Robertas Damasevicius
多核处理器在数据中心得到了广泛的应用。在多核处理器上运行的应用程序可能会因共享资源的争用而受到性能干扰，这可能会对在线应用程序的QoS产生负面影响并减少收入。为了保证在线应用的服务质量，数据中心往往会为在线应用过度配置资源，导致大量核心处于空闲状态，导致资源利用率极低。提高资源利用率的同时保证在线应用的服务质量是数据中心面临的一个挑战。以前的大部分工作都集中在固定状态模式下的干扰预测，这影响了其在生产数据中心的有效性。在本文中，我们提出了一种新的干扰预测方法，即dCCPI-predictor，它动态预测跨核心的性能干扰的多个应用程序一起运行，以确定“安全”的共址共享服务器资源。dCCPI-predictor为每个应用建立干扰预测模型，从而能够计算应用在任何共址中遭受的性能下降。dCCPI-predictor动态地适应应用程序的状态变化，预测不同状态下的性能干扰，这在以前的工作中被忽略了。我们在模拟数据中心上进行了多个基准测试，以评估我们的方法。结果表明，dCCPI预测器可以预测性能干扰，具有很高的精度，这是大大优于静态方法上级。
Multicore processors are extensively adopted in data center. Applications running on multicore processors may experience performance interference due to the contention for shared resources, which can negatively affect the Qos of online applications and reduce revenue. In order to guarantee the QoS of online applications, data center always over-provision resources for online applications, leaving a large number of cores idle, resulting in extremely low resource utilization. Improving resource utilization while ensuring the Qos of online applications is a challenge issue for data center. Most of the previous work has focused on interference prediction in fixed state mode, which affects its effectiveness in production data center. In this paper, we propose a novel interference prediction approach, namely dCCPI-predictor, which dynamically predicts the cross-core performance interference of multiple applications running together so as to identify the ’safe’ co-locations to share the server resource. dCCPI-predictor builds an interference prediction model for each application that enabling calculate the performance degradation that the application suffers in any co-location. dCCPI-predictor dynamically adapts to the state change of the application, predicting the performance interference in different states, which was overlooked in previous work. We conducted experiments on a simulated data center over multiple benchmarks to evaluate our approach. Results show that dCCPI-predictor can predict performance interference with a very high accuracy, which is greatly superior to static approach.
登录
查看更多内容
影响因子：
6.8
作者：
Eduardo Sontag
通讯作者：
Eduardo Sontag
DOI：
10.1145/3210560
发表时间：
2018-08
期刊：
ACM Transactions on Architecture and Code Optimization (TACO)
影响因子：
--
作者：
Yannis Sfakianakis;C. Kozanitis;Christos Kozyrakis;A. Bilas
通讯作者：
Yannis Sfakianakis;C. Kozanitis;Christos Kozyrakis;A. Bilas
DOI：
10.1145/2465351.2465388
发表时间：
2013-04
期刊：
--
影响因子：
--
作者：
Xiao Zhang;Eric Tune;R. Hagmann;Rohit Jnagal;Vrigo Gokhale;J. Wilkes
通讯作者：
Xiao Zhang;Eric Tune;R. Hagmann;Rohit Jnagal;Vrigo Gokhale;J. Wilkes
DOI：
10.1145/2000064.2000099
发表时间：
2011-06
期刊：
2011 38th Annual International Symposium on Computer Architecture (ISCA)
影响因子：
--
作者：
Lingjia Tang;Jason Mars;Neil Vachharajani;R. Hundt;M. Soffa
通讯作者：
Lingjia Tang;Jason Mars;Neil Vachharajani;R. Hundt;M. Soffa
DOI：
10.1145/1944862.1944887
发表时间：
2011-01
期刊：
--
影响因子：
--
作者：
Jason Mars;Lingjia Tang;M. Soffa
通讯作者：
Jason Mars;Lingjia Tang;M. Soffa