Responding to the sudden increase in computing demand, is cloud explosion just a bubble?
Although the concept of cloud explosion sounds good, there are still many problems in transferring the sudden increase of computing demand to the public cloud.
Because the fees are paid only when they are actually used, cloud eruption has aroused many people's interest in public cloud computing, including senior management and IT managers facing budget constraints. The hybrid cloud can transfer the load elasticity to the public cloud when the demand is peak, and reduce the size when the demand is reduced, and run on the local server. In this way, large enterprises can obtain guaranteed unlimited and uninterrupted services without paying high fees.
It's easy to understand why cloud bursts are so attractive. In the hybrid cloud model, the load can be seamlessly migrated between different clouds and dynamically adapt to changing needs. This is a typical case where the IT department responds to the needs of the business department in a timely manner. However, in fact, it is not only a headache to let the sudden increase of local load use the public cloud resources that are relatively close and unlimited. For most enterprises, this is impossible.
"I don't think this way will be as popular as everyone expected," said Edward Haletky, chief analyst of The Virtualization Practice. "If you want to achieve this degree of automation, enterprises need to absolutely trust the automation mechanism and implement appropriate monitoring and control. Can all companies do this? I know some companies can, but they are very large companies. "
What challenges does cloud explosion face
Bob Plankers, virtualization and cloud architect of Midwestern University, said that although the concept of cloud burst is very attractive, there are still many complex problems in the process of extending local applications to public clouds. These challenges even exceed the advantages of cloud eruption, so it may be a more efficient way to purchase several new servers in some environments.
"There may be serious network latency between the local computing environment and the public cloud," Plankers said. "Many enterprises have not evaluated the Internet outlet bandwidth before considering this approach. Latency will cause slow application response, which is the most important problem that enterprises should avoid."
What exactly is a cloud eruption?
Although the ultimate goal of cloud explosion has been widely agreed - to deal with the temporary rapid growth of computing demand - many experts and analysts still have different opinions on its necessity. Experts agree that the sudden increase in local load can be migrated to the cloud environment. However, they do not believe that the migration goal can only be public clouds (or private clouds), and the migration and return processes are automatic. Finally, if the enterprise can successfully deal with the problem of peak demand, it is not important how to call this process.
Most enterprise traditional applications do not support seamless scaling into the cloud. Considering this situation, if the accounting department of the enterprise is making year-end settlement, it will definitely lead to the peak load of the ERP system. IT administrators may consider allocating a new front-end engine to handle this demand, but they still need to communicate with the back-end SAP system, said Keith Townsend, the basic architect of AbbVie, a drug research and development company.
"If the data is still stored in the local data center, then using a new engine in the cloud will not play any role," Townsend said. "Relatively speaking, creating new virtual machines in private data centers will be a simpler and cheaper way."
The network is one of the other major challenges to deal with cloud bursts in a modern way, but this challenge is not limited to transmission delay and connection speed. If the application runs between the local server and the public cloud at the same time, the security will become more complex, Townsend said.
"(If in the local data center) I can set two policies in the two firewalls of the network layer to monitor the traffic between the two Web servers. But if I transfer the suddenly increased load to the cloud, who will ensure security and who will implement logging?" He said, "What should I do if I conduct an external audit?"
Even if the IT department can meet the challenges of infrastructure, cloud explosion will still bring a lot of management and business problems. In many cases, the improvement of performance is often accompanied by the increase of complexity, Haletky said.
"Many people talk about this wonderful story, but I don't know how many enterprises actually use cloud eruption because you must have a very high degree of automation," Haletky said. "How many companies can achieve that degree of automation? Netflix can, but can all companies be like Netflix?"
If cloud burst is adopted, it needs to plan in advance, that is, deploy virtual machine images and data in the cloud in advance, so that the public cloud can take over part of the load in time when the application has a peak load. This is not an unsolvable problem for businesses that can predict future demand, but it is difficult for administrators to decide how much performance the public cloud needs to provide and how much they are willing to pay for these ready data.
Ironically, one of the original purposes of using cloud eruption is to reduce expenses and costs, but in the end it backfired.
"If I automate a task, but do not let it close automatically after completion, it may lead to an increase in overall cost," Haletky said. "The overall idea to deal with the load explosion is to run when needed and return to the default state after everything is normal. If you buy 20 new Web servers from Amazon, you will incur huge expenses every hour."
Plan for unstable applications
The simplest and usually the cheapest solution to deal with the sudden increase in computing demand is still the traditional way, which reserves enough server performance locally. Experts said that this way does not necessarily mean buying new servers. The first step to evaluate whether the sudden load needs to be transferred to the cloud is to evaluate the current resource usage, Townsend said.
"In many cases, what we should focus on is not how to migrate the sudden load to the cloud, but how inefficient the existing work efficiency of the enterprise is." Townsend said, "Let's evaluate the existing resources, and then reasonably adjust the resources to have enough performance to complete all tasks."
For enterprises that encounter large-scale demand changes or lack performance, using automated tools such as Puppet and Chef, together with virtualization and cloud management platforms, can help them build a flexible hybrid cloud. However, there are many new manufacturers trying to use unique ways to reduce the complexity of cloud bursts.
Ravello Systems helps enterprises map application dependencies, manage network configurations, and clone the entire application stack into the cloud. However, most users do not choose to clone and migrate the burst applications in the production environment to the public cloud, but adopt a different way, said Shruti Bhat, product marketing director of Ravello.
"We see that users still place production environment applications in their own data centers while migrating loads from other environments, so that they can reserve enough performance and space for the production environment, such as migrating development and test environments to the public cloud," Bhat said, "This is still a way of cloud eruption, but in doing so, you will no longer have to worry about performance problems and the security risks brought by connecting applications through VPN and storing data in other locations."
Ravello's customers include 888poker, an online game company, which migrated QA and certificate environments to the public cloud. Two large banks also used Ravello's products to run the cloning environment of the network security laboratory. When the local performance reaches its peak, Red Hat also uses Ravello's products in its OpenStack training courses to provide students with a multi server experimental environment. Bhat said, "Red Hat found that every time students log in, they need to allocate a fast established OpenStack environment. If five to six servers are allocated to each student, the data center resources will soon be exhausted. When students log in, they will let students access the server in the data center, and when the resources in the data center are exhausted, they will start to let students access the OpenStack resources provided by Ravello. "
