Optimizing Multi-deployment on Clouds by Means of Self-adaptive Prefetching

Abstract

Abstract. With Infrastructure-as-a-Service (IaaS) cloud economics getting increasingly complex and dynamic, resource costs can vary greatly over short periods of time. Therefore, a critical issue is the ability to deploy, boot and terminate VMs very quickly, which enables cloud users to exploit elasticity to find the optimal trade-off between the computational needs (number of resources, usage time) and budget constraints. This paper proposes an adaptive prefetching mechanism aiming to reduce the time required to simultaneously boot a large number of VM instances on clouds from the same initial VM image (multi-deployment). Our proposal does not require any foreknowledge of the exact access pattern. It dynamically adapts to it at run time, enabling the slower instances to learn from the experience of the faster ones. Since all booting instances typically access only a small part of the virtual image along almost the same pattern, the required data can be pre-fetched in the background. Large scale experiments under concurrency on hundreds of nodes show that introducing such a prefetching mechanism can achieve a speed-up of up to 35% when compared to simple on-demand fetching.

Scheduling Jobs in the Cloud Using On-Demand and Reserved Instances

“…Moreover, for the same machine configuration, the clouds offer different billing options on-demand-, reserved-, and spot-instances, which are charged differently. Scheduling enough resources to meet user demands yet keep the cost low while adapting to workload changes remains challenging, despite recent research efforts [9][10][11].…”

“…We assume that clouds have infinite capacity. Each newly provisioned VM needs serval minutes to be booted [10,14]. An VM is charged per hour; even a factional consumption of less than one hour is counted as one hour.…”

Leveraging Collaborative Content Exchange for On-Demand VM Multi-deployments in IaaS Clouds

“…Using a decentralized storage solution (such as a parallel file system [13][14][15] or a dedicated repository [16]) reduces contention thanks to striping, but is only partially effective in our case, because the VM instances often access the same chunks in the same order. In our previous work [7], we show how to alleviate this issue by means of adaptive prefetching, however I/O contention to the repository is still a potential problem for scalability.…”

“…Since the VM instances of multi-deployments often follow a similar access pattern, a natural idea in this context is to enable the VM instances to talk to each other and "help" each other out in order to reduce the pressure on the remote repository. Based on the observation that I/O contention leads to jitter [7] (i.e., slight differences in time when the same chunk is accessed), we propose to organize the VM instances in a peer-to-peer topology where each VM has a set of neighbors, with whom it "gossips" about the chunks that should be fetched on-demand. Based on this information, VMs are able to anticipate future trends in access pattern and obtain chunks from their neighbors before they are actually needed, effectively preventing costly remote accesses if the anticipation was successful.…”

“…To achieve this, we leverage the fact that the VMs of the group typically have highly similar access patterns (e.g., during the boot phase they access the same chunks of the virtual disk in the same order [7]) in order to build a collaborative scheme where VMs exchange chunks between themselves in anticipation of expensive concurrent I/O accesses to the remote repository that would follow, which thus can be avoided. Our approach can dynamically adapt to the access pattern: it increases the rate of exchanges when remote accesses were successfully avoided, and backs off when the success rate starts dropping due to diverging access patterns.…”