Stay-Away, protecting sensitive applications from performance interference

Abstract

While co-locating virtual machines improves utilization in\ud resource shared environments, the resulting performance interference\ud between VMs is difficult to model or predict. QoS\ud sensitive applications can suffer from resource co-location\ud with other less short-term resource sensitive or batch applications.\ud The common practice of overprovisioning resources\ud helps to avoid performance interference and guarantee QoS\ud but leads to low machine utilization. Recent work that relies\ud on static approaches suffer from practical limitations due to\ud assumptions such as a priori knowledge of application behaviour\ud and workload.\ud To address these limitations, we present Stay-Away, a\ud generic and adaptive mechanism to mitigate the detrimental\ud effects of performance interference on sensitive applications\ud when co-located with batch applications. Our mechanism\ud complements the allocation decisions of resource schedulers\ud by continuously learning the favourable and unfavourable\ud states of co-execution and mapping them to a state-space\ud representation. Trajectories in this representation are used\ud to predict and prevent any transition towards interference of\ud sensitive applications by proactively throttling the execution\ud of batch applications. The representation also doubles as a\ud template to prevent violations in the future execution of the\ud repeatable sensitive application when co-located with other\ud batch applications. Experimental results with realistic applications\ud show that it is possible to guarantee a high level of\ud QoS for latency sensitive applications while also improving\ud machine utilization.Peer ReviewedPostprint (published version

Cloudy in guifi.net: Establishing and sustaining a community cloud as open commons

“…We found limitations in performance isolation required to avoid resource interference across containers, to be able to ensure the quality of the services provided. We developed solutions [28,29] but these are not yet integrated.…”

“…Although we found this isolation acceptable in terms of security and privacy, we identified resource isolation and admission control as an area of improvement to ensure the quality of the services provided (see Section 2). We did research and proposed solutions [28,29,53] for performance isolation, including resource prioritisation and pinning, elastic resource provisioning, reactive tuning, and admission control, but any of these are included yet in our software. Given the wide range of essential and optional software and services (IaaS, PaaS, and SaaS) for a decentralised CC, Cloudy can have a similar role in terms of standardisation and unification at the cloud infrastructure and service level as the guifi.net website has had at the network infrastructure level.…”

“…There are other BFS-like cloud services offered by large commercial providers, 28 like Dropbox, Google Drive, SpiderOak, Box, SugarSync. Most of them offer a freemium service, with a limited amount of free storage space available to engage customers into one of the paid subscription plans offering more space.…”

The Unobservability Problem in Clouds

“…Such an approach might not be feasible for customers, such as financial companies, who wish to protect their workloads. Stay-Away [46] regulates the performance of a time-sensitive application by proactively throttling other co-located batch applications. However, the authors assume that the class of the application running on a user VM is known, and that the VM performance can be throttled.…”

Measuring performance degradation of virtual machines based on the Bayesian network with hidden variables

“…However, studies like Ongaro et al, 2 Somani and Chaudhary, 3 and Padala et al 4 have reported that the current implementation of virtual machine monitors could not provide effective performance isolation mechanisms among VMs, and thus, the performance interference cannot be eliminated completely. It has been pointed out in previous studies [5][6][7][8][9] that the VM performance is influenced quite a lot by co-located VMs running simultaneously. Performance interference due to physical resource competition may lead to VM performance degradation, whose extent is closely dependent of the characteristics of VM runtime environments, such as the type of the applications running on other VMs and the number of VMs running at the same time.…”