Citation Context

...rent schemes of partitioning using simulations. While simulations are important and necessary when designing new architectures, examining largescale applications on existing hardware is important for improving our understanding of emerging workloads as well as improving performance in the deployed systems. Effective scheduling approaches to addressing resource sharing on SMT processors are also proposed [39, 13, 35]. In the area of OS and runtime scheduling, most studies focus on job co-scheduling to avoid co-locating cache contentious applications together to improve performance and fairness [25, 46, 14, 6, 31, 21, 20]. Approaches to alleviating resource contention and guaranteeing QoS by controlling the execution rate through hardware features or a runtime are proposed [18, 29]. Banikazemi et al. [4] present a scheduler to adaptively schedule threads to cores to take advantage of the cache topology to alleviate resource contention. Most of the above works use single threaded applications and focus on the resource contention aspect. Tam et al. [44] present a technique to cluster communicating threads onto the same chip to reduce the communicating latency. However, their approach focuses on the constructive ...

Citation Context

...in Section 4, the hypervisor can monitor the VMs between processes and attempt to schedule those workloads that do not conflict. This approach, often applied to multicore and multithreaded scheduling =-=[6, 10, 29]-=-, detects workloads with conflicting resource usages via statistics and processor performance counters, and attempts to schedule them at different times, so they do not concurrently share the contende...

Citation Context

...ne), and applications are frequently re-written and updated. Other approaches estimate the resource usage of applications and attempt to schedule applications with complementary needs together ( [3], =-=[5]-=-, [7], [10], [11], [15], [22], [24], [28], [36], [38], [42], [49]). While some general predictions can be made about application performance, it is challenging to make such predictions precise in the ...

Citation Context

... technique and build a sufficiently broad control strategy to realize adaptivity for a vast class of applications. Related to methodology, cloud research in general relies either on analytical models =-=[8, 16, 64, 73, 74, 76]-=- or on running actual experiments and building empirical traffic profiles and signatures [28, 52, 65, 67, 68]. Our system uses an analytical model to infer performance from measurements taken from the...

Citation Context

... individual critical sections. Gang scheduling. Chameleon’s cluster scheduling is similar to gang scheduling in that it will try to schedule multiple CPUs simultaneously. Past work on gang scheduling =-=[45, 27, 16, 6]-=- focuses primarily on time-sharing gangs of threads. While Chameleon can do this, there is little benefit because performance improves by running time-sliced threads in parallel on separate cores. Thu...

Citation Context

... been done to explore optimal task scheduling in different parallel platforms [19], [20], [21]. Especially, in AMC, many studies on scheduling focus on resource allocation at the OS level [22], [23], =-=[24]-=-, [25]. They aim to achieve high system throughput by balancing the hardware resources (e.g., cores, caches) among different programs. In [26], several phase co-scheduling policies are proposed for th...

Citation Context

... and I/O resources. In terms of this model, a MapReduce scheduling algorithm was developed to reduce the impact of interference among applications and to preserve task data locality. Bhadauria et al. =-=[9]-=- proposed a resource-aware co-scheduling algorithm that was capable of improving overall performance of applications and reducing power consumption. The key idea of this algorithm was to co-schedule a...

Citation Context

...d applications running on multi-core based systems. B. Consolidation and Resource Management One line of work on consolidation and resource management targets to optimize multiple nodes through techniques, such as turning off idle cores and load balancing across server nodes [5]. At the single server node level, a deeper understanding of workload characteristics (i.e., other than utilization) is required to reduce resource contention and improve energy efficiency of the server nodes. Bhadauria et al. propose co-scheduling algorithms based on bus and cache accesses of consolidated applications [6]. Authors propose coscheduling policies that manage allocated time and space share of applications to improve energy efficiency. Their proposed approach relies on offline energy-delay measurements to make co-scheduling decisions. Dhiman et al. propose a VM migration technique based on application characteristics to improve the energy efficiency [7]. Their proposed technique identifies the workloads that have complementary characteristics and coschedule them on the same physical resources. However, their proposed technique assumes equal resource allocations for coscheduled applications. These r...