Efficient resource provisioning in compute clouds via VM multiplexing

Abstract

Resource provisioning in compute clouds often requires an estimate of the capacity needs of Virtual Machines (VMs). The estimated VM size is the basis for allocating resources commensurate with demand. In contrast to the traditional practice of estimating the size of VMs individually, we propose a joint-VM provisioning approach in which multiple VMs are consolidated and provisioned together, based on an estimate of their aggregate capacity needs. This new approach exploits statistical multiplexing among the workload patterns of multiple VMs, i.e., the peaks and valleys in one workload pattern do not necessarily coincide with the others. Thus, the unused resources of a low utilized VM can be borrowed by the other co-located VMs with high utilization. Compared to individual-VM based provisioning, joint-VM provisioning could lead to much higher resource utilization. This paper presents three design modules to enable such a concept in practice. Specifically, a performance constraint describing the capacity need of a VM for achieving a certain level of application performance; an algorithm for estimating the aggregate size of multiplexed VMs; a VM selection algorithm that seeks to find those VM combinations with complementary workload patterns. We showcase that the proposed three modules can be seamlessly plugged into applications such as resource provisioning, and providing resource guarantees for VMs. The proposed method and applications are evaluated by performance data collected from about 16 thousand VMs in commercial data centers. The results demonstrate more than 45% improvements in terms of the overall resource utilization.

Composable architecture for rack scale big data computing

“…Various workload analysis demonstrated that 30~50% (and sometimes up to 90%) of server memory capacity goes unused [52][53][54][55]. Many of these scenarios indicated that CPU capacity is exhausted before memory capacity is reached, therefore leaving a fraction of the memory unused.…”

Joint Computing and Electric Systems Optimization for Green Datacenters

“…Authors in [27] present a control-oriented model that considers cyber and physical dynamics in datacenters to study the potential impact of coordinating the IT and cooling controls. To achieve further power savings while maintaining the QoS level, joint relationships among VMs, like load correlations, have been exploited in recent works [36,24,19]. For instance, in [24], Meng et al proposed a VM sizing technique that pairs two uncorrelated VMs into a super-VM by predicting the workloads.…”

Resource management of replicated service systems provisioned in the cloud

“…The most widely used traces are World Cup web site workloads that date back to 1998 [Arlitt and Jin, 2000;Petrucci et al, 2011], or are derived from the TPC-W benchmark [Singh et al, 2010], which was last updated in 2001. In contrast to conventional approaches, we seek an alternative to generate the workload -converting the CPU utilization traces of an existing production system into the workload input of a discrete simulator [Verma et al, 2007;Chen et al, 2005;Meng et al, 2010]. According to the basic utilization law [Kleinrock, 1975], the utilization multiplied by a normalized constant is essentially the request rate, especially when the load is below 100% utilization.…”

“…Following approaches used in [Verma et al, 2007;Chen et al, 2005;Meng et al, 2010], we adopt the utilization traces from current IBM production systems as workload input for each service, i.e., to generate the invocation requests. Based on the basic utilization law [Kleinrock, 1975], the utilization multiplied by a normalized constant reflects the request rate, especially when the load is below 100% utilization.…”