The "Cloud" most often refers to the Internet seen as the network of all networks. Cloud Computing is a new and delocalized means that uses the Internet for both designing computing resources and consuming them. It should be realized that using the Internet to access those resources is a major change. Understood properly, the Internet is an abstraction that designates a worldwide interconnection of computing resources. With Cloud Computing, everything happens as if computing resources were not localized geographically anymore. Moreover, users themselves adopt an ever-more itinerant way of working and therefore they would like to access information and applications through many different terminals, whether fixed or mobile, and from anywhere: at home, in a cyber cafe, in train stations, airports, and even within the company's offices!
The abstraction is indeed not only with respect to geography, but also with respect to technical infrastructure. Most Cloud Computing players (like Google, Amazon, eBay, and Salesforce) keep their technical infrastructure and their applications secret. There are two main reasons for these secrecy habits. The first one is obvious and is related to the security of platforms that are open to the Internet. They are inherently more vulnerable to attacks than closed platforms. The second is related to the new architectures that are being set up. These differ notably from the more classical architectures. Therefore, companies investing heavily in such infrastructure will obviously want to keep secret their main competitive advantage. Setting up such a large-scale distributed architecture constitutes a very high barrier that competitors in the same market will have to overcome.
Another important point that characterizes Cloud Computing is its new billing model. Customers are now billed "on demand", which means that they will pay only for those resources that they actually use. The costs related with producing those services and maintaining or expanding the associated hardware and software architecture to meet the demand, are now entirely on the side of the service supplier.
In summary, we can say that Cloud Computing is a new mode for delivering and consuming computing resources. These are delivered as services that can be used seamlessly from anywhere, because they use only the most basic standards of the Internet (TCP-IP, HTTP/HTTPS).
The outsourcing of computing resources to specialized providers is, by itself, not a novel idea. It has been applied for many years under the name "ASP", which stands for "Application Service Provider". But looking more closely, you realize that ASP and Cloud Computing are really two very different approaches to outsourcing. Now we'll look at why.
ASP providers were historically hosting providers that offered an alternative to in-house hosting to their customer. In this approach, the ASP offer the use of applications built by other vendors (for example, hosting of a MS Exchange or IBM Notes email solution for the larger telecom companies). This new separation of responsibilities between the software designer on one hand, and the company in charge of the production on the other, necessarily induces complications such as upgrading to the latest version, providing patches, and resolution of incidents. It also weakens the autonomy of the customer. Moreover, when a vendor claims to provide a cloud version of an existing solution, you should be extremely cautious.
Very few of them actually have the needed expertise and the scale that could allow them to offer low operating costs. Most of the time, such claims are not much more than a "cosmetic operation" cooked up by the marketing service to boost sales of existing software. In ASP mode, the customer usually pays for a package and later adds options when the demand for storage capacity increases.
Cloud providers are pure players that offer their services in a hosted mode. They are simultaneously vendor and provider of their own solutions (for example Google Apps, SalesForce, Amazon, AWS). In general, there is no such concept as a version of the deployed software. Moreover, adding features happens seamlessly and continuously for the customer, unlike traditional software, which evolves stepwise. Billing is on a pay per use basis and it is therefore quite easy to change those resources (and hence costs) without the hassle of renegotiating a contract.
The most common use of ASP is the deployment of a standard edition of some software on hardware dedicated to a single client. In this approach, each client has its own dedicated servers hosted by its ASP provider. At first glance, it might seem reassuring to keep separate resources for different customers. However, this automatically leads to poor equipment usage, significant additional costs, and low flexibility when the traffic load and the need for processing power suddenly increase. Moreover, some applications are not compatible with cluster deployment and don't take full advantage of such a topology. This limits their processing capacity.
In Cloud mode, there is a thorough separation between an application and the hardware on which it runs. Customers now buy a certain amount of processing power, a number of accounts, or some storage space rather than dedicated servers. To implement this, the inner architecture of the solution must have been designed properly. This implies that an application should be able to leverage any additional processing capacity. Moreover, the application should also be designed from the start to allow managing all customers on a single platform. This is often referred to as a "multi-tenant" architecture. As already mentioned, this explains why only new players, who have developed applications specifically for this model, will be competitive in this market.
To analyze the various hosting modes, we will categorize computing resources into four distinct layers:
- The hardware: This includes technical premises (white rooms, data centers), servers and networking equipment such as routers, proxies, or firewalls.
- The middleware: This includes essential software components that are necessary for running applications efficiently. Databases, application servers, and EAI belong to this category.
- Software: In this category, we have business applications such as ERPs, customer relation management (CRM) tools, or messaging tools. Most often, these applications reside on a platform that includes one or more middleware components.
- Operational resources: These include human resources and processes that handle IT operations. All common operations on a product platform such as deployment, upgrading, backup, and restoration belong here.
Now let's look at the four main hosting modes that we need to distinguish.
In this mode, which is often referred to as "on-premise", a company manages all the components of its applications directly. It owns its server premises where all equipment running the middleware and the applications is installed. Local teams are in charge of daily operations and deployment.
IaaS is an implementation of the Cloud Computing services on the lower layers of the IS. It allows renting processing units and storage capacity on demand. IaaS implies that the middleware and IT operations remain under the responsibility of the company that uses the IaaS.
PaaS can be seen as an extension to IaaS that goes beyond the rental of hardware. Namely, it includes middleware elements such as databases and application servers. PaaS can be used for instance by a traditional software vendor who would like to propose its products in SaaS mode. PaaS can also be used to run internal applications, by which we mean developed in-house. Unlike IaaS, the middleware is now under the responsibility of the service provider. Software and IT operations both remain under the control of the company that buys the services.
This is the most advanced version of Cloud Computing. It offers software packages as services rather than as traditional executable software. There is a profound difference between SaaS and classical software. SaaS is nothing less than a set of ready to use services that requires no installation and no maintenance whatsoever for the company that uses them. This constitutes a radical change in the way a company operates its IT resources.
A wide range of hosting modes for enterprise applications is currently available. The general trend, however, is towards outsourcing of computing resources and it is fair to say that Cloud Computing largely contributes to this movement. The following figure recaps the various hosting modes and areas of responsibilities that follow:
Before we go any further in our study of SaaS solutions, let's recall some basic facts concerning enterprise software architecture. The SaaS model is actually not suited for all architectures and you should be aware of the limits that apply when SaaS is used jointly with older architectures.
C entralized architectures are the most ancient architectures used in ISs. They are based on the simple principle of a passive terminal connected to a central system through the network. The latter manages absolutely everything, the session, the user interface, the business processing, and the storage of data. Two types of centralized architectures can currently be found in companies. On one hand, there are mainframes and machines such IBM's AS 400s. On the other, there are virtualization systems for the desktop, like Citrix or VNC. The following figure illustrates this kind of architecture:
Centralized architectures are definitely not compatible with the SaaS approach for several reasons. First, they don't rely on modern and open standards. Next, they require permanent connections. Finally, they are not built on the server side using flexible, multi-tenant architectures.
Cl ient-server architectures were massively deployed during the 90s, when relational databases and desktop computers became available. They are all based on client software (called the thick client) installed on the user's desktop and permanently connected to a database through a local network. The following figure illustrates this kind of architecture:
The client-server architecture isn't compatible with SaaS either. Here is why:
- Using a permanent connection is a bottleneck on a resource that is actually critical for any server side application
- Using a thick client implies many issues related to its deployment (installation, upgrade) as well as a strong dependency on the desktop
- The protocols for accessing databases are all proprietary and will be blocked by firewalls
Web architectures rely mainly on a web browser (which is by now available on any operating system) and a set of standards like HTTP and HTML to access web applications. These protocols have many advantages, including standardization. They allow a better use of networks and server-side resources when compared to connected and proprietary protocols used by client-server architectures. The disconnected nature of the HTTP protocol implies less resource consumption than that required for maintaining a permanent connection in the case of client-server architecture. A single server can thus handle the requests from a large number of users.
Web architectures are well-adapted to SaaS constraints and nearly all SaaS applications rely on this kind of architecture, whose flexibility and openness allow quick deployments with low coupling to the desktop system.
By s tandalone architectures, we mean applications that were designed specifically for the desktop, running Windows most of the time and sometimes Mac OS. They don't use any network connections. These applications run autonomously on the desktop. Business productivity tools like Microsoft's Word, Excel, PowerPoint or MS Project, all belong to this category.
So far, these applications have resisted the wave of web architectures and thus, logically, the SaaS model as well. They indeed require a level of usability that HTML pages have had a hard time matching. The situation is however about to change for several reasons:
- The upcoming version of HTML (version 5) will allow designing interfaces that are as sophisticated as those traditionally designed for Windows or Mac OS desktop applications.
- The recent availability of the Google Apps suite has drastically changed the situation in the office tools market. These had actually changed very little over the preceding years. The historical dominance of Microsoft's Office suite is now likely to be challenged in the near future. Google has the means to offer very low costs and unique collaborative features.
- The number of companies that try to greatly reduce the cost of their IS, their desktops, and their office suites is increasing exponentially.
A "private cloud" ref ers to the application of some cloud computing technologies within a given company. The wording first appeared when some companies decided to reuse internally the methods and the tools of cloud computing.
In fact, many IT specialists consider that, strictly speaking, this wording is misleading or even meaningless. The delocalization of computing resources is what really makes the essence of Cloud Computing. In such conditions, referring to a private cloud is simply an oxymoron. Moreover, a central idea of the cloud is that each customer pays only for the resources that were really consumed which, here again, loses its relevance in the case where the services are hosted on-premise.
One last point: it should be clear that the robustness of the datacenters, like those of Google, largely stems from their size and their geographical spread. Few organizations, whether public or private, can rely on such large-scale deployments. But these scale effects are precisely the primary reason for the benefits of Cloud Computing and they also favor optimal energy consumption.
Despite the semantic criticisms that can be made regarding the relevance of the private cloud concept, it is nonetheless true that many companies are currently interested in high availability and scalability, and that these two features precisely characterize cloud computing. Current solutions are most often based on expensive implementations of clustering. They are often complicated to deploy and far from satisfactory. Henceforth, the new platforms use virtualization on a scale that was unknown until recently. This opens new horizons to IT management and it is therefore easy to understand that companies now try to master these technologies for in-house use. The IT departments that position themselves as internal service providers have adopted the "private cloud" wording. It should be clear though, that this is really just another form of the traditional virtualization, rather than genuine Cloud Computing. In the end, the private cloud is best understood as a marketing concept, rather than a technology or a real IT concept.
Except for larger companies, it is unlikely for a "private cloud" to reach the same consolidation level as the public clouds. However, the "private cloud" can be a real opportunity for IT departments who wish to virtualize business applications for which no public offer is currently available.