Making the data bigger and better means that it can bring up problems, which also apparently goes against the database community “for over two years” (Agrawal). Assigned database that is designed to system, the first generic solution, had to deal with data not being together in one machine. The machine that was not able to do much, than just a few things because of the crippling effect. Meaning it would cause failures and overheads, and because of that most of these systems were not used too much. Recently, the development of a different class of flexible data management systems, like google, yahoo and amazon and more (cloud Computing). All of these systems deal with petabytes (Kerstetter) of data, they serve an online request with tight and those that are availability.
One of the major contributing factors towards the profitable of these systems today is the data model supported by these systems, a collection of key-value pairs with consistent and good read and write operations only on single keys. Even though a hug fraction of these are web applications, they are forced to only a single key access. Such as, online multiplayer games, social networking, and many more require a high access beyond a single key. Resulting in a key value used today, stores cannot help give those applications and also count on the traditional database for storing other content. Able to expand good stores that drive the in house applications of the corporations that have designed these stores.
Since the demand of cloud computing is becoming bigger, many applications are moving towards it. The power of being flexible with your resources and the new method of “pay as you go” (cloud computing) many models have broken down the structure barrier to only a few applications, this can also be tested with large corporations. They go on and off occasional, with the increasing demand for data storage and availability all the time causes many new challenges for data management in the cloud. These new applications demand a system that is capable of providing reasonable and consistent data as a service in the cloud (Kerstetter). The first step in this is designed along the key- values stores like “big table” and “hace” (Kerstetter) which does not provide good access to multiple objects. On the other hand, relying on a traditional database in order, and results in a profitable bottleneck for these applications, making it hard to get and benefiting the cloud. Making the demand for data management systems, that can help make a “gab between key values and traditional database systems” (Kerstetter) bigger.
The goal is to make data management system flexible and better in order to maintain achievement and making sure it is available to larger data without many attachments. Resources being used should be “highly dynamic” (Agrawal).Has mention before that features of major systems are from google, yahoo, and amazon. The design choices for each one are not always from the point of view, sometimes the concepts that are used have nothing to do with the design. Careful analysis is necessary to improve everything later in the future. My goal for this paper is to analyze the different systems and getting down the main functions of the data management systems with application in the cloud.
The application itself is very important through this processes. The data that is assigned to the system are designed to go through large amounts of data for applications that do support the system. This means the “application has a specific data and application state” (Cloud). The application is a place where at least two or more orders are made and out of those two only important or larger make it out. “Everything is based on the needs of the applications” (cloud).
Data model and applications: The “key is a table where the items that where each product is a key in value pair” (Fitzgerald). The value can either be an uninterrupted string, or it can have a good stable structure, the fact is supported by the quality of a single item (Fitzgerald). The system can now provide the correct operations. Making sure the data has access to only one object, resulting in a bigger and much simpler design. This makes the designers a more flexibly in operating to a greater quality. “The application level data manipulation is restricted to a single compute node boundary and thus obviates the need for multi-node coordination and synchronization using 2PC and Paxos” (Fitzgerald).
Consistency of single object: The process is only restricted to one key, meaning it will provide one object making sure it: “scalability is tractable” (Fitzgerald). If there is no object being replicated, all request for an object goes to the hosts. Let’s say is the entire data is set to separate along the host, it would “key nature of the request is to make them limited to a single use” (Fitzgerald). The system can now provide operations like read and write.
Replication process: The system today needs to support per object replication for high availability and to improve the accomplishment, in providing a good and smooth transaction. “Different systems use different devices to synchronize the replicas. providing different levels of smooth consistent throughout” (Agrawal).
Availability: Most of the time database is considered the entire data as a whole, and “hence”, part of the data was considered as not available throughout the entire systems. but a single object of an application has allowed data to be less equate. As a result, the system can manage certain parts of the data they are not available, providing reasonable service to the rest of the data. A system today must be together according to “6 non availability of a single component of the system” (Technology in Action) making sure the system is unavailable to other. On the other hand, the system today is not put together, and the sometimes not availability to a certain portion where the system might not affect other parts of the system. An example would be if there was not a divider, then that that does not affect the availability of the rest of the system.
The disadvantage of cloud computing like security issues, privacy issues, availability and data problems will hopefully no longer be a serious management problem, researchers and organizations are always looking for solutions to this problem (Geng). “The success of this being the system should work smoothly and efficiently, while using original of the cloud to deal with irregular workloads of new technologies in the cloud and provide a many degrees of consistency and availability making sure that each application requirements” (Geng). The over view of data management systems has the ability to expand to store on one end, and flexible transaction but not to bring down the database on the other end. Providing good and efficient data management to the wide range of applications in the cloud that requires “bridging this 10 gap” (cloud) with a system that can provide different degrees (consistency and scalability). The goal of this paper was to lay out the structure of the design of a system controlling the cloud data.
Design choices are based on two concepts: the value of the structure, which means it shows the importance of one value than the other. Once a design decision has been made, it allows the large values to take over or overrule on the other values becomes harder. And once that happen PNUTS give a better “flat row like a stable structure” (Geng). A row a can be pretty large and changes can be made every once and a while. Row however, have many empty columns for webs workloads. Other options should be included, but limited to the number of columns. Allowing each column that contains a list or a mixed up structures. Roe designs are based on page size and hence for more flexibility.
The second design choice is storage not paired together, the data is attached to a separate server responsible for things in the data. Since the data is on its own it is responsible for the data inside each other, and this is can only happen when the store has the data and can also run the server at the same machine. However, storage that is not paired together between the storage and the server, this makes it very hard to get involved with much more data splitting. “This design will need some structure to make sure that servers are located as close as possible to the actual data for not being dependent on much more righted pairing” (Geng).