Application vs Infrastructure, who's ready?

"Every cloud company will have the network ready, then deploy 40 to 80 servers," said Ken Duda, a vice president of software engineering at cloud expert Arista NX. "In his view, to avoid this situation, you need to understand the new topology and network, and the network should be homogeneous, and be as automated as possible."

The first problem of cloud network design: The type of cloud

Glyn Bowden, a storage industry group SNIa, says companies must first understand the public, private, or mixed cloud. After understanding and implementing, most enterprises will eventually implement a hybrid cloud, which refers to the simultaneous use of both public and private components in the same system.

What are the advantages of this model? Once stored, the cloud is equivalent to implementing different layers of data in different clouds. At the same time, for business systems, this could mean mounting highly sensitive applications and data in a private cloud and deploying management or billing tools into a public cloud.

This choice is based on trade-offs between cost and use-case requirements. For example, using a public cloud for storage avoids buying and building the infrastructure, and it can also address backup requirements. However, if the enterprise needs to recover quickly, the required low latency connection will be expensive. At the same time, private clouds may require more construction costs, but these investments generally yield rewards such as flexibility and the ability to achieve optimal tight-flow management.

This decision may also vary depending on the application and workload types involved. "You need to add filters to your workload-first, economic, then trust, and finally technical," said Sandra Hamilton, vice president of EMC Consulting EMEA. She explained that workloads have different trust requirements, where "trust" includes speed, performance, security, and so on. In one workload, economic considerations may dictate traditional architectures, and for another workload, economic and trust based considerations may choose a public cloud. However, even now there may not be a public cloud service that meets the requirements, so the company may have to adopt a private cloud model.

Chris Rae, vice president of Cloud Solutions at the software company CA, suggests that, in general, many companies can "deploy in-house private clouds for testing and development." ”

The difference of cloud network topology

When it decides to adopt a private cloud or hybrid cloud, it often has to consider a broad, flat network that is better suited for data and applications to move horizontally between servers. This enables the team to adjust the application's use of processing power and bandwidth. In a traditional network, the data mainly flows to the core network, then to the downstream aggregation layer, which is called the North-South movement.

Localized Cloud network design

The need to increase speed and reduce network latency in order to meet service level agreements also means that more work is done locally. Johan Ragmo, a Nordic data business development manager for Lucent, points out that one of the solutions is to deploy an application service module in the switch, so that some tasks, such as security and load balancing, can be done in the rack.

This approach, in conjunction with the bundled approach of companies such as Lucent, Brocade, and Cisco, consolidates servers, storage, and switches into a previously tested, off-the-shelf component, rack, chassis, or container. The concept here is that once a standard component is connected to the automatic distribution layer, all included components are added to the cloud resource pool.

Another way to prevent delay and optimize transmission is to automate the network, not only to support resource allocation, but also to support network configuration management.

"To automate assignments and tools-computing, networking, and storage." You have to liberate people from these processes so that they can do more important work. and capacity is important, you need to automate capacity management and modeling, so you want to run at 90% to 95% of the load level, but do not reach 110%, "said CA's Rae."

Ultimately, cloud computing also has a golden ideal goal: to provide business applications and data that meet established service levels. Rae says that while the simplification and flattening of the network infrastructure is important, it's just a means to a goal, not a final goal. "In my opinion, the starting point is not a technical or hardware layer; This means that you need to provide gold, silver, copper and other different levels of large, medium and small services. He says the gold ratio is 80/20--80% of the demand is taking up 20% of the cost.