3.2. PROPUESTA TEÓRICA
3.2.6. Equipo de Trabajo (Organización y Funciones)
A variety of physical and architectural differences are found between cloud computing and the other computing systems just discussed. As with most IT system
concepts, however, these differences generally tend to be a difference of degree rather than type. Those who argue that cloud computing is just “the modern version of the timesharing model from the 1960s” (Schneier) miss the fact that one does not need a mainframe to enter the cloud and in fact, the Cloud came about because of the increased speed allowed by newer technology and the existence of the Internet. At the same time, the idea of connecting
multiple users to what appears to them to be a single source has been around for a long time. So, how can cloud computing be distinguished from other types of computing?
Scalability is primarily achieved in grids by increasing the number of working nodes, whereas in the Cloud it is achieved by automatic resizing of virtualized hardware resources (Vaquero et al.). Clouds can perform this more easily than grids because they typically tend to be centrally controlled and managed, whereas grids tend to have decentralized control and self-management. In addition, service level agreements and quality-of-service guarantees are an inherent feature of cloud computing, whereas they are generally layered on top of
architectures such as the grid, and entirely absent from cluster computing. Because cloud computing offers isolation of resources (as opposed to the Grid’s resource sharing), pricing becomes simpler for the consumer and more problematic for the service provider. The consumer sees a pay-per-use model; the service provider must assess the best approach for ascertaining appropriate pricing.
Although the kind of dynamic, runtime provision of resources seen in the Cloud could potentially be available in some grids (Čibej, Sulistio, and Buyya 2009; Kurdi, Li, and Al- Raweshidy 2008), batch processing has tended to be the norm for grid computing (Jones
2008; Foster et al. 2008). Also, according to several authors, within grids high-performance and throughput are achieved with both parallel and distributed processing arrangements that often use high-end computers, as opposed to cloud and cluster computing which tend to rely on commodity servers to reduce cost, albeit with a potential sacrifice of performance
(Armbrust et al. 2009; Buyya et al. 2009; Čibej, Sulistio, and Buyya 2009; Agrawal et al. 2010). However, Chetty and Buyya (2002) suggested that grids composed primarily of commodity machines do exist.
Vaquero and colleagues also noted that grids have historically been developed for scientific purposes. Scientific projects have typically received public funding for particular projects, leading to more centralized approaches to resource allocation. They are usually billed using a fixed rate per service or require different organizations to share idle resources, creating virtual organizations which are allocated resources on a fair use basis. They do this by using resource brokers to determine fair use on the basis of automated policies (Buyya, Abramson, and Venugopal 2005; Buyya, Shin, and Venugopal 2008; Murphy et al. 2010; Kotrotsos et al. 2010). Clouds, on the other hand, typically provide commercial services and “are usually billed using a pay-per-use model” (Vaquero et al. 2009, 54).
Although grids attempt to ensure a fair share of resources across organizations, clouds use virtualization to provide an illusion of a single dedicated resource. They do not rely on explicit sharing but rather, provide resource isolation through virtualization. Both grid and cloud support the aggregation of heterogeneous hardware and software resources, but grids typically virtualize data and compute resources whereas clouds typically also include hardware virtualization (Vaquero et al. 2009).
Grids offer such services as metadata search and data transfer, whereas that type of service is still underdeveloped in the Cloud (Vaquero et al. 2009). Nonetheless, when it comes to quality of service (QoS), grids tend to lag behind clouds, with service level agreements (SLAs) needing to be created via applications that reside on top of the grid. SLAs are an inherent feature of clouds (Vaquero et al. 2009). In addition, the overall service orientation of grids in comparison to clouds is somewhat different: “Grid computing
specifically refers to leveraging several computers in parallel to solve a particular, individual problem, or to run a specific application. Cloud computing, on the other hand, refers to leveraging multiple resources, including computing resources, to deliver a ‘service’ to the end user” (IBM 2009, 6).
Foster and colleagues (2008) note that the ability to manage and track provenance has typically been built into grids via workflow systems, and has also been built for grids as a standalone service, PreServ. They define provenance to be “the derivation history of a data product, including all the data sources, intermediate data products, and the procedures that were applied to produce the data product” (6). They add, however, “Provenance is still an unexplored area in Cloud environments, in which we need to deal with even more
challenging issues such as tracking data production across different service providers (with different platform visibility and access policies) and across different software and hardware abstraction layers within one provider” (2008, 7). At least one attempt to deal with
provenance in the Cloud has occurred, however. Muniswamy-Reddy and colleagues (2009, 2010; Muniswamy-Reddy and Seltzer 2010) describe what they call a Provenance-Aware Storage System (PASS) to augment the Amazon Web Service (AWS) for backend storage in order to provide the capability of tracking and managing provenance in the Cloud.
Table 1 provides a detailed comparison of cloud computing and three other
architectures discussed in the literature, (i.e., cluster, grid, and peer-to-peer), highlighting a variety of features.