Dalian: The construction and reflection of data center based on cloud computing

The construction and reflection of data center based on cloud computing technology has now become a panacea, no matter what solution will eventually be attributed to cloud computing. In the wave of cloud computing, should university data centers deploy cloud computing? How to make cloud computing real landing, and really play the benefit, not only a concept, but also a question worth pondering.

The practice of virtualization technology in data center

Since 2003, Dalian University of Technology has set up a centralized data center for the unified management of the servers for external information release from various departments in the school. The construction of the data center has gone through two stages: the first phase uses the independent 1U physical server to provide the service, the most has more than 200 physical servers. In this mode, the application of backup management, physical server fault handling, room area, power supply, refrigeration consumption and other issues are concentrated. Starting from 2006, the school began to test the server virtualization technology, and in 2007 years built the university's first large-scale VMware virtualization server farm, starting from the second phase, the original physical server application on the gradual migration to the virtualized server, now more than 280 virtual machines are running. Further expansion of the virtualized server cluster in 2009, the first virtualization servers were upgraded with memory and gigabit network adapters in 2010, and the data center switch, Cisco Nexus 7018, was deployed. At present, the data center has deployed 16 CPU virtualization server clusters, all physical servers with dual 10GbE network adapters, dual 4Gb FC HBA Allied, the physical server cluster has maintained a reasonable load.

In the process of server virtualization implementation, found that the physical server memory consumption rate is much higher than the CPU consumption speed, so in the server cluster expansion, the new server memory increased to 144GB, the original server memory also expanded to 64GB. Another important factor that affects the efficiency of server virtualization is the performance of the storage system. To further improve storage system performance, in 2011, we carried out a Fastcache technology upgrade to the existing storage system EMC cx4-960, using a multiple 100GB solid-state disk to cache the entire system, which made the storage read-write latency significantly improved.

Figure 1 Datacenter Virtualization server cluster topology

In the aspect of virtualization platform security, strict authority management and backup system are designed. For example, resource management has only the right to allocate virtual machines and configure designated networks within its own resource pool; backup is done in the background, and resource management cannot delete backups. On a backup system, EMC Avamar provides a unified global weight-reduction backup that can provide any one-week backup of any virtual machine in half a year. For critical servers, use VMware Data recovery for daily backups, and tape libraries for critical applications for application-level backups. Through strict management measures and multilevel automatic backup mechanism, the security of data center is ensured effectively.

Coordinate with internal organization to plan reasonable management structure, integrate the broad infrastructure (including network, storage, virtualization platform, backup, computer room environment, etc.) into one department management, and other departments become users of the department. Practice has proved that this model can effectively improve the utilization rate of resources, improve the efficiency of management, and initially formed the Department of Internal Infrastructure cloud service delivery.

In the process of implementation of the virtualization platform, the operation and management cost of the data center is greatly reduced, and the physical server of the data center has been reduced by nearly half from 200 units. The management department carries on the application promotion, the migration cost drops significantly. Through templates, snapshots and other mechanisms, management can easily deploy applications, when the application upgrade failure can be a complete fallback.

How to get cloud computing to the ground in the data center

In the process of data center virtualization, the feasibility and necessity of server virtualization in University Data center are proved, and the actual benefit is realized. So what's the relationship between cloud computing and virtualization? The relationship between the university's cloud computing and virtualization can be summed up as follows: The goal of cloud computing is to implement on-demand use of applications, virtualization only provides a means of resource pooling. Can only automate, providing a large number of virtual machines is not cloud computing. The core problem with cloud computing is not the pooling of resources, but whether the application can be truly seamless and scalable on demand.

Therefore, we think that the key factor that affects the implementation of the data Center cloud computing in colleges and universities is whether the information-related application is really oriented to cloud computing and can truly provide services in the cloud computing mode. Server virtualization or infrastructure virtualization is just the first step in the evolution of cloud computing. The university cloud computing deployment should be a gradual process, along with the understanding of cloud computing and application development, gradually evolved into cloud computing. The evolution process of cloud computing in Dalian University of Technology is located as follows:

We have just taken the first step in building a cloud-oriented data center.

The traditional application architecture does not consider the problem of dynamic expansion. For example, during the peak of the course, we want to automatically expand an application server dynamically to accommodate the number of requests, and automatically return to the usual number after the peak period. Now we can partially solve this problem by adopting server virtualization and application server clusters, but the process is not automated and still requires manual configuration and intervention. With the further improvement of application pressure, the traditional application server, middleware and database architecture can not guarantee the real dynamic expansion.

In the PAAs phase, we hope to explore the new application architecture, in the message communication mechanism, data access mechanism, automatic extension mechanism, etc., to achieve a real cloud-oriented development platform. On this basis, the new development of the application gradually migrated to the cloud computing platform, implementation of the application of on-demand use.

In the context of cloud computing, we think that in addition to adapting to the application architecture of cloud computing, the key technologies currently in the datacenter include:

1. Network equipment supporting data network and communication network integration

The cost of the current Wan is greatly reduced, and the onboard Wan Shao (LOM) will gradually replace the onboard gigabit. Mainstream onboard LOM has begun to support both FCoE, iSCSI Uninstall, and IP transmission. Mainstream storage vendors have started to support FCoE directly in high-end storage. High-end data center switches have started supporting specifications such as DCB, which can provide two-tier multilink, packet-free Ethernet, and other key features. Therefore, the full implementation of data network, communication network integration time is ripe.

2. Centralized storage devices that support FCoE and tiered storage

From our current experience of running and testing, the FC protocol is still the only choice for a highly reliable datacenter storage LAN. FCOE-supported switches, can interface cards are very mature, and storage devices have directly supported the FCoE interface. The technical risk of replacing traditional FC with FCoE is relatively small, and the traditional FC's operation and maintenance experience can still be continued. Hotspot data and inactive data must exist in large-scale virtualization implementation. Solid state disk prices are relatively high, the full use of solid-state disk in the cost is difficult to accept. Storage devices need to be able to dynamically differentiate hot-spot data, automatically migrate hotspot data to solid-state disks or higher performance disks, migrate inactive data to low-cost hard drives, and can deliver high-performance services at reasonable cost.

3. Network, security equipment, software, virtualization

In large-scale virtualization deployments, traditional physical networks and security devices are no longer applicable. Traditional security devices cannot be deployed to secure control between virtual machines. We believe that software-based virtual devices will replace some of the physical security devices. For example, a antivirus gateway between host and VM, a distributed software switch that controls security policy between VMS, a pure Software Web application Protection device, and a pure software firewall device.

4. Integration of communication networks and virtualized infrastructures

Traditional communication network control is based on three-tier routing control or two-tier access port control, in the virtualization infrastructure is no longer applicable, virtual opportunities on the two-tier port migration, access to the same host virtual machines between the physical switch. In a virtualized environment, physical switches should be able to perceive the migration of virtual machines and adjust control strategies dynamically. Physical switches should be integrated with virtualized switches to extend virtual switches as physical switches.

To sum up, the University Data Center Deployment cloud computing technology background is ripe, through the cloud computing technology deployment can effectively reduce the data center operation management costs. However, it is still a subject to study how to make the cloud computing technology fall to the ground completely. We hope to be able to use the cloud-oriented application development platform for research and deployment as an opportunity to fully cloud the application of information. There are some key technical issues in building a cloud-oriented data center, and we will focus on these key technologies in future data center construction.