Workload-Aware Provisioning in Public Clouds

Abstract

Public cloud services rely on virtualization to support multitenancy-customers from different organizations are allowed to share the data center infrastructure. Unfortunately, today's public clouds fail to provide sufficient isolation. Hardware resources are often multiplexed between virtual machines that belong to different customers, and they can cause performance interference to each other. This article characterizes the interference on an important metric, the network latency between virtual machines, and shows that Amazon's EC2 cloud, a leading public cloud provider, suffers from a long tail latency problem. The root cause of this problem is co-scheduling of CPU-bound and latency-sensitive tasks. We leverage these observations in Bobtail, a system that allows cloud customers to proactively detect and avoid these bad neighboring virtual machines without any help from cloud service providers.

Affective design using machine learning: a survey and its prospect of conjoining big data

“…The captured user perceptions were used to determine the optimal camera which satisfied both functional and affective design needs. Xu et al (Xu et al 2014) proposed to use interactive online big data to examine the realtime relationship between customers' affective satisfaction and facial expression, where the facial data was captured and analysed in real time. Their proposed framework attempted to differentiate between posed and spontaneous expressions but still cannot address all affective states other than facial expression.…”

On the Information Freshness and Tail Latency Trade-Off in Mobile Networks

“…The LTP of an update of a certain application is defined as the probability that the latency is greater than a certain threshold. We note that the tail probability of latency has been shown to affect the user experience much more than the mean [20], [21], and that motivates us to consider the tail metric.…”

Virtual resource pricing scheme in cloud platforms

“…Regardless the convenience of cloud platforms, unpredictable workload becomes the major problem when operating a cloud-based data centre (Chaisiri et al, 2012;Corradi and Rana, 2014). For example, a spike in the requested VMs may result in higher service rejection rates and experienced delays by users due to congested resources (Xu and Musgrave, 2014). The most mentioned solution to deal with this problem is to over-provision the virtual resources at all times so as to meet user's maximal demands (Mian et al, 2013;Xu and Musgrave, 2014).…”

“…For example, a spike in the requested VMs may result in higher service rejection rates and experienced delays by users due to congested resources (Xu and Musgrave, 2014). The most mentioned solution to deal with this problem is to over-provision the virtual resources at all times so as to meet user's maximal demands (Mian et al, 2013;Xu and Musgrave, 2014). Obviously, this solution will lead to low resource utilisation especially during the periods of low user demands.…”