Dynamic Memory Pressure Aware Ballooning

Kim, Jinchun; Fedorov, Viacheslav V.; Gratz, Paul V.; Reddy, A. L. Narasimha

doi:10.1145/2818950.2818967

Cited by 12 publications

(12 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To improve the memory efficiency in the hypervisor environment, the memory overcommitment technique that decreases or increases the memory allocated to VMs is used. Waldspurger [23], Zhou et al [24], Zhao et al [25], Guo [26], Kim et al [27], and Lu and Shen [28] periodically profiled the access frequencies of pages by nullifying the translation look-aside buffer (TLB) entries of randomly selected pages. Based on this, the least accessed pages are returned when the amount of memory is insufficient.…”

Section: Related Workmentioning

confidence: 99%

“…Although several studies [23][24][25][26][27][28][29] dynamically adjusted the memory allocation of existing VMs, in these studies, memory was taken from a specific VM and allocated to another VM when the physical memory was insufficient. As these approaches assume an environment in which each VM has an independent virtual address space, it is difficult to apply those techniques directly to the full GPU virtualization environment in which the same virtual GPU memory space is shared.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Ballooning Graphics Memory Space in Full GPU Virtualization Environments

Park

2019

Scientific Programming

View full text Add to dashboard Cite

Advances in virtualization technology have enabled multiple virtual machines (VMs) to share resources in a physical machine (PM). With the widespread use of graphics-intensive applications, such as two-dimensional (2D) or 3D rendering, many graphics processing unit (GPU) virtualization solutions have been proposed to provide high-performance GPU services in a virtualized environment. Although elasticity is one of the major benefits in this environment, the allocation of GPU memory is still static in the sense that after the GPU memory is allocated to a VM, it is not possible to change the memory size at runtime. This causes underutilization of GPU memory or performance degradation of a GPU application due to the lack of GPU memory when an application requires a large amount of GPU memory. In this paper, we propose a GPU memory ballooning solution called gBalloon that dynamically adjusts the GPU memory size at runtime according to the GPU memory requirement of each VM and the GPU memory sharing overhead. The gBalloon extends the GPU memory size of a VM by detecting performance degradation due to the lack of GPU memory. The gBalloon also reduces the GPU memory size when the overcommitted or underutilized GPU memory of a VM creates additional overhead for the GPU context switch or the CPU load due to GPU memory sharing among the VMs. We implemented the gBalloon by modifying the gVirt, a full GPU virtualization solution for Intel’s integrated GPUs. Benchmarking results show that the gBalloon dynamically adjusts the GPU memory size at runtime, which improves the performance by up to 8% against the gVirt with 384 MB of high global graphics memory and 32% against the gVirt with 1024 MB of high global graphics memory.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Ballooning Graphics Memory Space in Full GPU Virtualization Environments

Park

2019

Scientific Programming

View full text Add to dashboard Cite

show abstract

“…Modern platform virtualization offer sophisticated automatic memory allocation mechanisms to adjust the memory in use by the VMs they host, based on the current memory pressure regime , . Furthermore, other approaches (e.g., Liu et al, Schopp et al, Kim et al, Zhang et al, and Moltó et al just to name a few) aim at improving existing memory management techniques by overcoming their limitations and by reducing their overhead (and hence their impact) on the running VMs. However, all of these mechanisms are unaware of application performance and thus may lead to SLO violations, unless they are complemented by approaches like ours.…”

Section: Related Workmentioning

confidence: 99%

“…However, all of these mechanisms are unaware of application performance and thus may lead to SLO violations, unless they are complemented by approaches like ours. Also, some of these approaches are of limited applicability as they require either a para‐virtualized system or a modified version of the hypervisor.…”

Section: Related Workmentioning

confidence: 99%

“…Modern platform virtualization offer sophisticated automatic memory allocation mechanisms to adjust the memory in use by the VMs they host, based on the current memory pressure regime. 47,69,70 Furthermore, other approaches (e.g., Liu et al, 18 Schopp et al, 71 Kim et al, 72 Zhang et al, 73 and (b) linear feedback controllers are usually less effective to cope with nonstationary workloads than those based on fuzzy control. 19,21,22 In a certain related study, 19 the authors propose APPLEware, an autonomic middleware for the joint performance and power control of colocated web applications in virtualized infrastructures, based on adaptive neuro-fuzzy modeling and on model predictive control.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

FCMS: A fuzzy controller for CPU and memory consolidation under SLA constraints

Anglano

Canonico

Guazzone

2016

Concurrency and Computation

View full text Add to dashboard Cite

Summary Cloud providers (CPs) rely on server consolidation (the allocation of several virtual machines [VMs] on the same physical server) to minimize their costs. Maximizing the consolidation level is thus become 1 of the major goals of cloud providers. This is a challenging task because it requires the ability of estimating, in a resource contention scenario, multidimensional resource demands for multitier cloud applications that must meet service‐level agreements (SLAs) in face of nonstationary workloads. In this paper, we cope with the problem of jointly allocating CPU and memory capacity to (a) precisely estimate their capacity required by each VM to meet its SLAs and (b) coordinate their allocation to limit the negative effects due to the interactions of dynamic allocation mechanisms, which, if ignored, can lead to SLA violations. We tackle this problem by devising FCMS, a feedback fuzzy controller that is able to dynamically adjust the CPU and memory capacity allocated to each VM in a coordinated way, to precisely match the needs induced by the incoming workload. By means of an extensive experimental evaluation, we show that FCMS is able to achieve the above goals and works better than existing state‐of‐the‐art alternative solution in all the considered experimental scenarios.

show abstract