Section: New Results

Efficient VM management in clouds

Participants : Alexandru Costan, Alexandra Carpen-Amarie, Gabriel Antoniu.

Infrastructure as a Service (IaaS) cloud computing allows users to lease computational resources from the cloud provider's datacenter for a short time by deploying virtual machines (VMs) on these resources. This model raises new challenges in the design and development of IaaS middleware. One of those challenges is the need to deploy a large number (hundreds or even thousands) of VM instances simultaneously. Once the VM instances are deployed, another challenge is to simultaneously take a snapshot of many images and transfer them to persistent storage to support fault tolerance and management tasks, such as suspend-resume and migration. With datacenters growing rapidly and configurations becoming heterogeneous, it is important to enable efficient concurrent deployment and snapshotting that are at the same time hypervisor independent and ensure a maximum compatibility with different configurations.

We addressed these challenges by proposing a virtual file system specifically optimized for virtual machine image storage [19] . It is based on a lazy transfer scheme coupled with object versioning that handles snapshotting transparently in a hypervisor-independent fashion, ensuring high portability for different configurations. Large-scale experiments on hundreds of nodes demonstrate excellent performance results: speedup for concurrent VM deployments ranges from a factor of 2 up to 25, with a reduction in bandwidth utilization of as much as 90 % [18] . We implemented this deployment scheme in the Nimbus cloud and presented a demo illustrating it at the Grid'5000 School [26] .

Given the dynamic nature of IaaS clouds and the long runtime and resource utilization of scientific applications, an interesting use-case for the multi-snapshotting techniques is for efficient checkpoint-restart. We introduced an approach that leverages VM disk-image multi-snapshotting and multi-deployment inside checkpoint-restart protocols running at guest level in order to efficiently capture and potentially roll back the complete state of the application, including file system modifications. This framework is specifically optimized for tightly-coupled scientific applications that were written using a message passing system (in particular MPI) and need to be ported to IaaS clouds. Our solution introduces a dedicated checkpoint repository that is able to efficiently take incremental snapshots of the whole disk attached to the virtual machine instances, thus offering support to use any checkpointing protocol that can save the state of processes into files, including application level mechanisms, where the process state is managed by the application itself, and process-level mechanisms, where the process state is managed transparently at the level of the message passing library. Experiments on the G5K testbed show substantial improvement for MPI applications over existing approaches, both for the case when customized checkpointing is available at application level and the case when it needs to be handled at process level.

We integrated this checkpointing scheme inside the Nimbus cloud with some promising preliminary results. We plan to complement the existing solution with live incremental snapshotting using asynchronous background transfers for high checkpointing efficiency and with adaptive prefetching to achieve high restart efficiency.