Cloudy with a Chance of Load Spikes: Admission Control with Fuzzy Risk Assessments

Abstract

Abstract-Elasticity is key for the cloud paradigm, where the pay-per use nature provides great flexibility for end-users. However, elasticity complicates long-term capacity planning for cloud providers as the exact amount of resources required over time becomes uncertain. Admission control techniques are thus needed to handle the trade-off between resource utilization and potential overload. We define a set of admission control algorithms that combine risk assessment methods with a fuzzy aggregation framework. An experimental evaluation using a mixture of bursty and steady applications demonstrate that our algorithms can increase resource utilization by a factor of two while limiting overload problems to a few percent.

Mining association rules for admission control and service differentiation in e‐commerce applications

“…Both internet server resource management and web data mining have attracted extensive research in the literature. Internet resource management can be achieved through resource provisioning (Urgaonkar et al, ; Urgaonkar, Kozat, Igarashi, & Neely, ; Villela et al, ; Xue et al, ), AC (Ashraf et al, ; Elnikety et al, ; Khojasteh et al, ; Konstanteli et al, ; Tomas & Tordsson, ; Wu et al, ), service differentiation (Chandra et al, ; Dutta et al, ; Hjort et al, ; Lakew et al, ), and request scheduling (Zhu, Yang, Chen, Wang, & Yin, ). It is common practice to combine more than one of the techniques to achieve better results.…”

“…ISPs always aim at maximizing their profits by making the best use of their resources, that is, while maintaining the SLAs, they will try to allocate resources to as many e‐commerce companies as possible. In system overloading situations, unless e‐commerce companies are willing to pay for extra resources, ISPs usually apply admission control (AC) Schemes (Ashraf, Byholm, & Porres, ; Elnikety, Nahum, Tracey, & Zwaenepoel, ; Khojasteh, Misic, & Misic, ; Konstanteli, Cucinotta, Psychas, & Varvarigou, ; Tomas & Tordsson, ; Wu, Garg, & Buyya, ) to their services. By rejecting less important requests when the system is overloaded, AC schemes can guarantee the performance of internet services.…”

Reducing Noisy-Neighbor Impact with a Fuzzy Affinity-Aware Scheduler

“…In our case, to account for this uncertainty and possible prediction errors, we presented a fuzzy admission control [11] that decides whether a new VM can be accepted without relying on user information about how tolerable their applications are to overbooking. The steps performed are: (1) predict the future status of the data center; (2) evaluate the possible impact of accepting a new VM by using a fuzzy logic engine that accounts for the uncertainty about future events [7]; and finally (3) take the admission decision based on the acceptable level of risk (threshold) of the datacenter at the current time. The risk thresholds are obtained through a distributed set of PID controllers that adjusts their own risk threshold based on current vs. target utilization levels (for more details see [11]).…”

“…Due to the complexity of the problem, as well as the uncertainty about future status of CPU cores and application needs, we take a fuzzy logic approach. Fuzzy logic is a highly expressive, natural, and efficient way to deal with uncertainty and approximate reasoning, and we previously applied this approach to cloud admission control with good results [7].…”

Cloud Service Differentiation in Overbooked Data Centers

“…Admission control decisions are based on a fuzzy logic risk assessment [19], combined with a proportionalintegral-derivative (PID) controller [20]. The PID controller changes the acceptable level of risk over time and the associated overbooking pressure, depending on the deviation of the current data center utilization from the target [1].…”