Cloud computing is not inherently unreliable, but like all forms of IT, cloud services must be carefully selected and managed to achieve specific reliability and availability goals. These steps can be in the form of a contract, in the form of technology, or even require rethinking the architecture of your application. If not carefully considered, the benefits you gain from cloud computing may be less than you expect.

SLA reduces the risks arising from the use of cloud vendor data centers

The first step to avoid being affected by cloud outages is to assess the reliability of the cloud vendor's data center. Most cloud service providers have a very small number of data centers, usually only one, and these data centers are prone to the same types of failures as enterprises. The most well-known cloud computing failures are often those events where the entire cloud computing data center malfunctions, usually caused by natural disasters. To protect yourself in the event of possible failures, you must request specific data center configuration information or obtain availability guarantees from your supplier.

For the availability of servers, storage devices and networks, the best strategy is to determine the SLA during negotiations, so as to specify the availability guarantee and the time to restore services in case of failure. It is very important for users to know whether disastrous weather such as hurricanes or snowstorms frequently occur in the area where the cloud computing data center is located. At the same time, it should also be determined whether the data center is equipped with a backup power supply and whether it is equipped with a backup data center that can take over normal operation.

The backup data center must be located in a different area from the main data center so that it will not be affected by the same problem. At the same time, it must also have sufficient capacity to handle the failover of cloud applications. Since few suppliers can provide sufficient backup data center capacity for 100% failover of the primary data center, the SLA will explain how to manage failover.

In this case, it might be necessary to pay for priority. If your cloud services include geographical diversity to support distributed user groups, then your own various facilities can provide certain protection measures against cloud vendor failures. Carefully examine your contract to ensure there is sufficient capacity to handle the additional load.

Network performance issues or deficiencies will lead to cloud downtime events

The most common cause of cloud computing failures is usually not the cloud computing itself, but the network. Most cloud applications are accessed via the Internet, and Internet availability is the culprit behind most cloud downtime incidents. The only way to solve this problem is to adopt virtual private networks or virtual local area network services, or ensure that multiple Internet Service providers (ISPs) provide services simultaneously for websites to access cloud applications. If the issues of security and compliance can be resolved and confirmed by the supplier's contract, then this is a very good choice. Unless the cloud provider has already used the private network service provided by the operator, it is very likely that you will need to pay a special fee.

As the cost of Internet services for small businesses continues to decline, it becomes possible to provide two ISPs for a branch office. However, please ensure that there are no common failure points between the two offices. Typically, peer points and shared interconnection "hotels" can be shared among multiple suppliers. Even the most common access cabling between ISPs may nulliven the benefits of a dual-network connection.

The elasticity problem of cloud applications must be solved

If both the cloud computing data center and the cloud computing network failure problems have been solved, then the next issue is the elasticity problem of the application itself. The biggest problems in managing high availability and cloud services both involve database access and reliable transaction processing.

If one data center malfunctions, even if another backup data center can back up the applications that use this data, the data stored in the data center will be unavailable. Unless the application data is maintained in a "hot standby" state at multiple locations, a single failure will lead to the loss of data access, thereby rendering most of the other redundant measures ineffective. This problem also exists in internal data center backup. Therefore, those enterprises that provide redundancy for their own data centers will find that the same measures will be equally effective in cloud computing. This is more of a financial strategy than a technical one. Due to the storage and service costs of cloud computing, the cost of maintaining redundant data in cloud computing is higher. A better solution is to deploy all your internal data in a highly available and protected data center and access it from multiple cloud computing locations.

The best usability management must be integrated with the application itself. At any time, database updates are carried out simultaneously for multiple replicas. If a failure occurs during the update process, there is a risk of data integrity being lost. Online transaction processing systems typically include a "two-phase commit" process to ensure that no problems occur due to the failure to successfully update all database copies. Sometimes, the update of a single database may also be in an uncertain state due to network failures. It is necessary to review applications developed specifically to ensure network failures or data center failures, so as to ensure that the stored databases do not have the risk of data corruption or inconsistency.

It is unreasonable to expect that cloud applications can have the same or higher reliability than internal applications. Furthermore, the reliability and specific goals you set might cost you a lot. When building your business case, please remember to consider the reliability cost, or you may find that your application has to make some compromise between reliability and cost.