撰文: 陈兆辉 新意网集团有限公司 科技总监
According to Structure Research’s recent estimation, the hyperscale cloud infrastructure worldwide market has just crossed the US$30 billion mark and the projected CAGR for 2017–2022 is 48.8%. Asia Pacific is definitely one of the most important and fast-growth playing fields that the hyperscale cloud providers will eye on to seize the market share. Amazon Web Services (AWS) has announced the opening of its Hong Kong region this year and Microsoft Azure is already hosting its East Asia region in this highly connected gateway to China. Hong Kong is poised to become a key location for these global cloud service providers. In order to stay competitive, colocation operators in Hong Kong should get ourselves ready when designing new data centers to meet the changing needs of these hyperscale cloud players.
High Power Density
Cloud computing has transformed today’s IT infrastructure. As servers are virtualized and provisioned in a more agile manner, the infrastructure supporting today’s data centers has also changed. The significant increase in power density in the past few years has brought about some major changes in the design of data center in this cloud age. Cloud service providers are requiring 6kW to 10kW per rack on average, with individual racks consuming as high as 20kW of power per rack.
Designing for a high power density environment cannot be an afterthought. Retrofitting an old infrastructure with a 3kW per rack design to support this cloud-level power density is at best inefficient if not infeasible. The cooling infrastructure has to be sized to support enough airflow and cooling capacity. There needs to be containment and return plenum design to handle the cooling requirements, which in term dictates the architectural design of the data hall.
Design for Cost-effectiveness
With resilience at data center level across availability zones or regions, global cloud providers are now satisfied with a N+1 design for UPS power. Simply put, if one data center becomes unavailable, the workloads in that data center will be shifted to another connected data center, thereby reducing the need for a 2N design for UPS power that have been a common requirement in the past decade to ensure high availability of data centers. Compared with a 2N design, an N+1 design saves cost and space for data center operators and helps achieve a more cost-effective solution.
Efficient energy design is also a major selection criterion as every decimal point in the PUE (Power Usage Effectiveness) may translate into millions of dollars per year for a multi-megawatt deployment.
Accommodating Heavy Racks
As server racks of major cloud players are pre-installed with loads of optimized servers that are extremely heavy, the data center needs to meet certain basic physical requirements, such as high ceiling, heavy floor loading, covered loading bays, and ramp-less delivery routes with high-capacity freight elevators.
Connectivity is Key
Low-latency connection among availability zones or regions is now key to the cloud architecture. Data center connectivity has therefore emerged as one of the top selection criteria. Carrier neutrality, availability of dark fibers in diversified routes, secure private trunks, and the presence of subsea cables have all become key requirements when selecting colocation data centers.
No Compromise in Security and Physical Risks
Physical security is undoubtedly a key consideration for cloud service providers who have done an excellent job in convincing their customers or potential customers that public clouds are now very secure. They therefore cannot afford to reside in a facility that does not meet their stringent security requirements or one that poses significant physical risks. Those that are close to the waterfront or in the proximity of hazardous facilities would probably fail the initial screening.
Trust on Operating Experience
Maturity in operation and service excellence are also key attributes of data center operators who serve these global cloud players. Proven track records in building and operating sizable data centers and the ability to meet stringent service levels are musts. An open Building Management System (BMS) that can be integrated with the cloud player’s monitoring systems has also become increasingly important.
Plan ahead for Future
Even if a data center meets all of the above requirements, the facility still may not be the choice of cloud service providers if it cannot deliver the needed infrastructure in time or lacks the flexibility for future growth. With an average annual growth rate of over 40% per year for the top cloud players, time to market and flexibility for growth are not just nice to have, they are a must to react to market demands. Everything being equal, the one that can provide ample growth capacity and a short delivery lead-time will win the deal. Good capacity planning and modular design are therefore key winning strategies.
The opening of a new AWS region in Hong Kong and the phenomenon growth of Microsoft’s cloud business will definitely fuel cloud adoption in Hong Kong. Let's get prepared for more cloud-ready data centers to meet the demand in the coming years to help build Hong Kong as a technology hub in the region.
点击此处以了解更多关于新意网的 AWS Direct Connect 解决方案：AWS Direct Connect