Automatic Scalable VM Management for Data Centre Optimisation

Periodic Reporting for period 1 - FishDirector (Automatic Scalable VM Management for Data Centre Optimisation)

Période du rapport: 2015-10-01 au 2016-03-31

The growth in the individual size and number of data centres has resulted in a relative shift in the energy footprint of the IT sector from devices to data centre and networks. The aggregate electricity demand of the cloud (including data centres and networks, but not devices) in 2011 was 684 billion kWh (if compared with the electricity demand of individual countries in the same year, the cloud would rank 6th in the world)5 and is predicted to increase by 81% by 20206. The energy costs of powering data centres is thus becoming a significant operational cost for data centre operators and has focused attention on improving the energy efficiencies of data centres and reducing their carbon footprint.
Whilst reductions in the individual energy consumption of servers due to new low power CPU architectures and multi-core designs have been readily implemented, there has been little improvement in the efficiency of server utilisation. As ‘on-demand’ services impose varying loads on a data centre, most servers will either be operating at a fraction of their capability or may even be unused yet still powered up. A 2008 survey by McKinsey & Company found that server utilization rarely exceeded 6% and
overall data centre utilization was as low as 50%7 whilst an NDRC study found that the average US server operates at only 5% -
15% utilization level while consuming 60% - 90% of its maximum system power8. A more recent IBM study indicated that despite hardware improvements, the mean server CPU utilization is still only 18% In order to reduce their hardware (i.e. server) costs and reduce their overall energy consumption data centres have begun to adopt virtualization, running multiple virtual machines (VMs) on each server. In a virtualized environment, live VM migration capabilities to relocate VMs both within single (and for large enterprises, across multiple data centres) can be exploited to achieve various resource management objectives, in particular reducing the number of active servers needed at any given time to meet a load (‘resource-load matching’) thus allowing inactive servers to be powered down (typically 15% - 30% of the servers
in a data centre are running but without meeting any of the load10). In addition virtualization eases server maintenance provisioning and fault tolerance management11.
In a modern data centre, the VM workload is dynamic, varying in response to user demand. However, data centre operators do not currently have a comprehensive automated system for optimally placing VMs and allocating the optimal number of servers to meet the workload requirements. This results in sub-optimal hardware utilization as mentioned above, high hardware (servers and storage) and facility CapEx, a high OpEx (energy consumption) and a poorer end user experience (lower response
time and/or intermittent availability). The few existing tools for managing data centre workloads do not scale, forcing data centres to be sub-divided into management entities of circa 100 servers, leading to greater operational complexity.

The Specific Objectives of FishDirector Innovation Project – Our FishDirector software has been developed to the stage
where it is ready for testing under real world conditions. We therefore propose a Feasibility Study to verify the technical and
economic viability of its implemtation including a technology state-of-the-art review to assess potential competitors, market
analysis, identification and planning for testing/demonstration, and formulation of an elaborated business plan, with a view to
continue to an application to phase 2 of the SME Instrument. We therefore propose to conduct a phase 1 Feasibility Study
with the following Objectives:
F1 - Market and competitor analysis:
• Assess user needs, demand and market segments and sizes.
• Identify purchase processes, channels and gatekeepers.
F2 - Cost assessment:
• Assess the product development, production and demonstration/testing costs.
F3 - Business plan development:
• Determine the optimal go-to-market strategy.
• Decide upon likely revenue stream(s) arising from implementation of the strategy.
• Develop the price model and structure.
• Conduct risk assessment.
• Formulate an exploitation and dissemination plan.
• Explore the possibilities and implications of product adaptation or customisation for third-party customers.
F4 - IPR analysis and novelty verification/network patent analysis (see section 2.2).
F5 - Demonstration design:
• Establish the requirements (hardware, personnel) for undertaking demonstrations of FishDirector at a data centre
under real load operating conditions.
F6 - Plan for phase 2 development
• Formulate a detailed work plan for the phase 2 project.

Sardina Systems has implemented a feasibility study on the commercialization of FishDirector technology, a technical and economic validation FishDirector software solution, including data and documentation ready for commercialization across European and US markets.

Substantial market and competitor SWOT analysis was conducted to contrast 7 major competitors, analyzing Sardina Systems as a company and the FishDirector product. The outcome highlighted the uniqueness of FishDirector and the extensive commercial opportunity available.

We carried out a full business analysis and business plan on commercialization of FishDirector technology. As part of this effort, we further refined our financial, marketing, technical development, pre/post sales services and human resources plan for further product development, customer support, operation and marketing. The cost estimates have also been validated include reasonable expenditures for maintenance, equipment, hosting and other development and operational costs.

Following an IPR analysis, we have also concluded on an IPR strategy focussed on enabling FishDirector to be rapidly brought to market in a competitive manner.

To enable Sardina Systems to both build sales and to develop next generation product, we have also the combined technical facilities capable to demonstrating product capabilities, as well as having adequate variation in environmental and workload input, to demonstrate key values of FishDirector without performance degradation.

We also had conversations with a number of major global cloud operators and global system vendors, who gave us feedback confirming the depth of cloud data center operators' challenges.

Building on the feedback and input from major global cloud operators and global system vendors, we have formulated a list of activities, detailed with commercialization and product refinement plans.

The outcome of the feasibility study is that Sardina should pursue the business opportunity.
Our solution is highly attractive to the market, as seen by the tremendous market pull we are experiencing pre-commercialisation. The activities to be completed prior to commercialisation are mainly related to scalability and should be fast-tracked. Having automated process will allow us to on-board clients faster and obtain high market penetration. Sardina will seek pursue this business opportunity through an SME Phase 2 proposal.

The ability to optimize the installation and configuration of multiple VMs running on multiple servers requires two key technical challenges to be overcome:
1) The Initial Placement Problem. This is the problem of optimally and rapidly placing VMs on multiple servers within a data centre with a large set of highly dynamic workloads.
2) Rebalancing Problem: This is the problem of live rebalancing the workload across the servers in a data centre while meeting workload requirements and minimising the number of servers needed to meet the workload. Our ambition in the FishDirector project is to create a highly innovative software solution that overcomes these challenges and works together with OpenStack to provide virtualized data centre management and automation. We wish to to transition FishDirector from its current TRL6 to TRL9 by demonstrating in a large data centre under normal operating conditions. Our solution will allow data centre operators to automatically optimize VM instances across multiple servers with no practical limit to the number of servers that can be managed. No existing solution exists that is able to do this. FishDirector software incorporates unique technologies, including the decision engines’ constraint solvers, a super-scalable data consolidation system, and an overall architecture designed to scale. It solves at scale the Initial Placement Problem and the Rebalancing Problem to dramatically lower operational expenditure and hardware and facilities capital expenditure, and maximize hardware utilization, ensuring high service reliability, flexibility and lower complexity. Improvement Potential of FishDirector - ICT products typically have a very short refresh cycle and are continuously being developed and upgraded to incorporate new technologies. At SDN we have planned the following developments:
• Leverage FishDirector’s deployment and configuration component to enable:
- Automated super-scalable patch/updates.
- Version controlled data centre wide configuration, reversible changes, merge changes.
• Leverage FishDirector’s unique Condition Engine to enable flexible user-programmed operations optimization.
• Leverage FishDirector’s flexible decision engines to improve storage, network I/O efficiencies.
• Expand coverage and capabilities of FishDirector’s super-scalable monitoring, log management and data warehousing systems.

Data Centre Requirements - The needs of data centre operators have been identified as:
• Minimising energy cost by increasing overall data centre energy efficiency.
• Minimising hardware (servers, storage) and infrastructure costs by optimizing server utilization.
• Achieving higher capacity at constant Total Cost of Ownership (TCO).
• Automating management operations and tasks (minimising ‘down time’, reduce labour costs).
• Automating workload migration in the event of faults (minimising ‘down time’ and maintain service integrity).
Server hardware and data storage account for approximately 67% of the power consumption of a typical data centre12 and are the major operational costs. Whilst low power CPUs can help reduce their power consumption, running multiple VMs on servers is the more effective way of increasing energy efficiency and can be readily implemented on legacy as well as new server hardware, thus extending their useful lifetime and maximising return on investment. It is estimated that virtualisation has a 65% penetration of the server market. By enabling the automatic consolidation of applications on fewer servers VMs running FishDirector can potentially improve energy efficiency by 10% – 40% by freeing up servers (which can thus be powered down) and releasing cooling capacity (the second largest power consumption, 29% of the total) for expansion or to provide redundancy14. Currently a typical data centre has a server capacity utilization of approximately 18% or less, and data centres worldwide in 2013 consumed 40 GW of power (7% more than in 2012)15. FishDirector will enable an increase in server utilisation to at least 40% and, if implemented across the entire global data centre sector it could reduce power consumption by 5.9 GW.

Efficient clouds using FishDirector

Periodic Reporting for period 1 - FishDirector (Automatic Scalable VM Management for Data Centre Optimisation)

Partager cette page

Télécharger