SYNERGY

Motivation

Facebook operates 5 data centres including 180000+ servers. Every day data increase ~ 500 TB; Number of everyday user accesses ~ 1 million.

According to IBM, aggregated losses in the data center energy supply chain can reach up to 97%. Approximately, the only one of every 27 Watts is spent directly on performing user tasks (considering the average CPU utilization is up to 20%).

Virtualization of computing resources allows to significantly reduce energy costs with more dense placement (consolidation) of virtual machines on the physical servers. Sun Microsystems and Emerson Network Power have reported that increasing the average load of a physical server from 10% to 70% (by increasing a number of virtual machines hosting on it from two to eight) will reduce the total number of servers in a data center by four times and will decrease the total energy consumption almost by three times.

Nevertheless, the consolidated deployment of virtual instances in practice leads to a decrease of their performance and stability of operation. Besides, it increases a risk of the appearance of a so-called ‘Noisy neighbor’ – a virtual machine, located on the same server, whose applications are aggressively using system resources (CPU, memory, disc or network I/O operations). Existence of a ‘noisy neighbor’ can lead to a significant performance degradation of the applications running on other virtual machines, and even cause a suspension of their operation.

Moreover, an effect called ‘software ageing’ can mitigate the problem. This phenomenon refers to software tendency to fail, degrade performance and increase system power consumption after running continuously for a certain time. It happens mainly due to memory leaks, memory and disc fragmentation, data corruption, numerical error accumulation, etc.

Project concept and architecture

Energy efficiency, performance and dependability of applications running in data centers should be optimized at different levels – from developing energy - efficient applications to optimal VMs allocation, dynamic tasks scheduling and proactive software rejuvenation enabling synergistic coexistence of virtual instances.

The purpose of the project is to develop a framework for improving dependability, performance and energy efficiency of synergistic data centres.

The project structure:

1. Software power consumption measurement and tariffication framework; this framework will provide a useful tools for software power effectiveness measurement, prediction and optimization. Besides, it will introduce a new tariffication model based on measuring power consumed by a particular service or virtual instance.
2. Energy-aware software rejuvenation strategy; we will examine power ageing phenomenon related to virtual instances and particular software services running on them. As a result we will develop a proactive strategy for software power rejuvenation (via VM reboot or services restart) allowing to reduce amount of energy consumed by a particular VM or a service, increase stability of operation and prevent performance degradation.
3. Smart tasks scheduling and virtual instances allocation for synergistic coexistence. The framework will provide tools aiming at sustainable coexistence and effective consolidation of virtual resources by optimal task scheduling and allocation of virtual instances taking into account system resources utilization profiles of neighbouring VMs. This will allow to effectively reduce the overall power consumption without affecting performance and dependability of different VMs by avoiding noisy neighbourhood.

Target Audience

1. Companies owning private or public data centres.
2. Cloud customers looking for stable Cloud-environment for deploying own services.

Sources of revenue

1. Decreasing electricity bills paid by DC owners and Cloud providers.
2. Providing more flexible tariffication model for cloud customers taking into account energy consumed by virtual instances.
3. Improving clouds and data centres computing and energy efficiency

Implementation stage

The research group runs extensive experiments proving the concepts of software power ageing and rejuvenation technique and sustainable co-existence of virtual instances.