Abstract: The skyrocketing demand for a new generation of cloud-based consumer and business applications is driving the need for next generation of datacenters that must be massively scalable, efficient, agile, reliable and secure. Based on an analysis of the Intelligent Networks in telecommunications to identify proven concepts and key lessons that can be applied to enable next generation IT.
Cloud computing experience this paper asserts that:
• In order to scale cloud services reliably to millions of service developers and billions of end users the next generation cloud computing will have to follow an evolution similar to the one that led to the creation of scalable telecommunication networks.
• In the future network-based cloud service providers will leverage virtualization technologies to be able to allocate just the right levels of virtualized compute, network and storage resources to individual applications based on real-time business demand while also providing full service level assurance of availability, performance and security at a reasonable cost.
• A key component - identified in this paper as the Virtual Resource Mediation Layer (VRML), must be developed through industry collaboration to enable interoperability of various public and private clouds. This layer will form the basis for ensuring massive scalability of cloud infrastructure by enabling distributed service creation, service delivery and service assurance without any single vendor domination.
• The next generation virtualization technologies must allow applications to dynamically access CPU, memory, bandwidth and storage (capacity, I/O and throughput) in a manner similar to that of the telecommunications 800 Service Call Model1 with one level of indirection and mediation. The next generation cloud evolution is a fundamental transformation, which will enable global service collaboration networks utilizing optimally distributed and managed computing, network and storage resources driven in real-time by business priorities.
INTRODUCTION:
Everyone has an opinion on what is cloud computing. It can be the ability to rent a server or a thousand servers and run a geophysical modeling application on the most powerful systems available anywhere. It can be the ability to rent a virtual server, load software on it, turn it on and off at will, or clone it ten times to meet a sudden workload demand. It can be storing and securing immense amounts of data that Is accessible only by authorized applications and users. It can be supported by a cloud provider that sets up a platform that includes the OS, Apache, a MySQLdatabase, Perl, Python and PHP with the ability to scale automatically in response to changing workloads. Cloud computing can be the ability to use applications on the Internet that store and protect data while providing a service - anything including email, sales force automation and tax preparation. It can be using a storage cloud to hold application, business, and personal data. And it can be the ability to use a handful of Web services to integrate photos, maps, and GPS information to create a mash up in customer Web browsers. By Anology with the telecommunications network, this layer will:
Ø Mediate between networked applications and virtualized computing, network and storage resources with dynamic provisioning;
Ø Enable development of end-to-end or application to- spindle Fault, Configuration, Accounting, Performance and Security (FCAPS) management
Ø based on business priorities using dynamic monitoring of workloads on computing, network and storage resources and;
Ø Allow the development of next generation converged service creation, delivery and assurance infrastructure that is massively scalable and globally interoperable along with a new degree of agility.
CLOUD FORMATION:
Animoto – a small startup with limited resources, created an online service that generates a unique custom video from photos and music uploaded by users. When they put the application on Facebook and it went viral and demand shot up through the roof. They are managed to scale from 50 servers to 3500 servers in three days – all without having to buy a single piece of hardware or having to create their own compute, network and storage infrastructure. This was all accomplished by renting compute infrastructure from cloud service provider.
For Example: Amazon Elastic Cloud Computing (EC2) and complementary service management capabilities from management provider Right Scale which enabled automated workload monitoring and Virtual Machine provisioning on Amazon’s EC2 infrastructure.
The above example demonstrates how existing cloud infrastructure can be used to enable massive scale and agility at a very reasonable cost using:
§ Virtualization technology to dynamically provision virtualized software applications, load balancers and web application servers on-demand,
§ Innovative distributed computing technology that allows database distribution,
§ A managed Service Oriented Architecture for Web Service deployment and
§ A large number of commodity hardware devices (servers, storage and network elements)
Impressive as it is, this current state-of-the-art in cloud computing still is just a baby step when compared to what is expected in a fully functional cloud based service creation, delivery and assurance platform.
Consider the following:
o While the infrastructure services used by service developers are dynamically provisioned, and billed on usage, the system administration and management costs continue to increase with the number of servers used.
o While service delivery is able to scale in the current cloud model to support spikes in demand, application availability, performance optimization and security management have to be implemented separately. Today, a host of other companies are actively trying to fill this need [2,3,4,5,6] with additional services using customized point solutions.
o Disaster Recovery (DR) and storage management (de-duplication, tiered storage) are mostly lacking and have to be individually implemented at additional cost and effort.
The above points highlight some of the reasons why the cloud is today divided into private and public instances. The rule of thumb that seems to have evolved is that if there is a need for developing and deploying services using more than 50 to 100 servers at near full utilization, then private clouds may prove economical. This is roughly the point at which the additional management cost and complexity required for service assurance – not just simple service delivery, makes private clouds viable. It is important to recognize that this number varies and depends on the extent of automation made available by the cloud infrastructure service providers to facilitate service creation, delivery and assurance. More the automation provided by public clouds, lesser the need for private clouds. History shows that economies of scale will favor public clouds if they can address availability, performance and security at all levels.
It is apparent that the datacenter infrastructure required to manage virtualized computing, network and storage resources in an integrated fashion has not yet evolved to take cloud computing to the next level. One of the reasons is that datacenters today are managed using a number of legacy management systems that invariably started with a server-centric management paradigm and have since evolved incrementally over the past couple of decades to accommodate the shift towards client-server and network based computing paradigms. As a result, there is no single system today that provides truly integrated cross-domain management capabilities required for a service-oriented cloud infrastructure. At best each management offers specialized management of a particular infrastructure silo (i.e. servers, storage and networks) or partial management across more than one silo. It is also quite common for similar management functionality to be duplicated in solutions provided by multiple vendors specializing in different domains . Further, the best practices promoted by each vendor may conflict when attempting end-to-end optimization across the datacenter.
To illustrate the above, take a look at any datacenter today and you are likely to find that they are paying thrice for a storage volume manager performing similar functionality in their servers, storage and network devices without even being aware of it. To ensure redundancy, clustering and multi-pathing may have been implemented in their servers, networks and their storage. Storage cache management is likely implemented in their virtual servers, physical servers and storage layers.
Figure 1 shows a typical datacenter with all its support systems demonstrating the incremental nature
of its evolution and the resulting complexity and cost.
Figure 1. Datacenter complexity today with duplication of management functions
Clearly the inefficiencies incurred in terms of management complexity, sub-optimal performance and costs are untenable. Dynamic reconfiguration of all infrastructure i.e. compute, network and storage resources, based on an application’s needs is a necessary condition for automating datacenter management.
For this paper we analyzed the IN services in telecommunications and propose that a similar evolution that utilizes dynamic provisioning of computing, network and storage resources made possible by virtualization technologies will radically reduce the management complexity in next generation datacenters. By borrowing the FCAPS management and signaling abstractions from the telecommunications domain, a next generation virtualized intelligent service collaboration network infrastructure can be developed that will integrate both public and private clouds to offer massive scale and interoperability.
Management simplicity can be achieved by consolidating application, server, network and storage management intelligence into the SCN and enabling the brokering of compute, network and storage resources between the various applications that need them based on real-time demands, workload profiles and business priorities.
THE CLOUD EVOLUTION:
(FAULT, CONFIGURATION, ACCOUNTING, PERFORMANCE AND SECURITY (FCAPS)MANAGEMENT AND THE INTELLIGENT SERVICE COLLABORATION NETWORK)
The current definitions of cloud computing are just beginning to incorporate end-to-end management as a basic foundation for cloud IT. For example, Forrester Research Group now defines cloud computing [5,9] as “A pool of abstracted, highly scalable, and managed compute infrastructure capable of hosting end customer applications and billed by consumption.” Meanwhile the ITU-T Telecommunication Management Network (TMN), already has a well articulated definition for managed infrastructure in the context of the telecommunications Intelligent Networks for voice services. In this layered model, each layer is responsible for different management functions, while interfacing with underlying and overlying layers, to provide a complete and comprehensive set of management capabilities:
1. The Network Element Layer (NEL) implements logical entities within a device
2. The Element Management Layer (EML), implements device level FCAPS management functions.
3. The Network Management Layer (NML), implements path management, topology management and fault isolation
4. The Service Management Layer (SML), implements mechanisms to assure service level agreements and ensure Quality of Service (QoS).
5. The Business Management Layer (BML), implements strategic enterprise management functions, such as budgeting and billing.
In this manner, the above TMN FCAPS framework enables:
1. Fault management, by detecting and correlating faults in network devices, isolating faults and initiating recovery actions.
2. Configuration management, by providing change tracking, configuration, installation and
distribution of software to all network devices.
3. Accounting management capability through comprehensive network usage reports generated by collecting and parsing accounting data
4. Performance management by providing real-time access for the monitoring of network performance (QoS) and resource allocation data
5. Security management by providing granular access control for network resources
Applying the above framework, we propose a Cloud Computing Reference Model that explicitly incorporates FCAPS management and defines the various roles of infrastructure, service creation, delivery, and assurance platform providers. These roles can be assumed by a single provider or multiple providers depending on whether the solutions are proprietary or standards-based.
However, history has consistently shown us that proprietary solutions may drive innovation initially but standards will ultimately be required to achieve massive scale by enabling the interoperability of competitive proprietary solutions.
Figure 2 shows the roles of various players (service operators, developers and end users) in order to realize massively scalable clouds where thousands of developers create millions of services that serve billions of customers.
A similar cloud model described by Frank Gillett is shown in Figure 3. However, that model does not seem to address end-to-end management. Ultimately, the cloud service infrastructure must provide end-to-end service assurance (FCAPS management) to meet both service creation and service delivery platform user requirements. The service creators must be able to develop services
Figure 2. Service Creation, Assurance and Delivery Model
rapidly using reusable and collaborating service components available globally. The infrastructure must also accommodate billions of users globally who will contribute to wildly fluctuating workloads.
Figure 3. Current Cloud Computing Model
Amazon has successfully demonstrated that virtualization, distributed computing and service oriented software environment can be combined with commodity hardware to both develop and deliver massively scalable services. It has created a virtual server environment that can be successfully used to create a certain class of applications (web based service delivery). Where it falls short is in the scalability of system administration. The end-user is left to worry about various datacenter functions such as load balancers, firewalls, replication, disaster recovery (DR), and storage and security management.
This has opened the opportunity for a host of startups to attempt to fill this gap.
Current cloud evolution is limited to the following three areas:
v The Virtualization of servers, load balancers, and some server IP address management services
v The replacement of SAN/NAS infrastructure with large commodity server farms that support virtual applications using Direct Attached Storage (DAS) or File Systems (distributed or otherwise)
v The Current approach to storage replication and storage based application management using multi-vendor SAN/NAS solutions is being made obsolete by the adoption of virtualization technologies. Next generation virtualization technologies will allow the network based IN services platform to utilize COTS storage elements which will be virtualized and dynamically allocated to provide the right throughput, IOPs and capacity to the right application based on business priorities.
v Application of distributed computing innovations through Web Services and Service Oriented Architecture (SOA).
It is apparent from above that the datacenter is evolving incrementally from the bottom up without the top down end-to-end architectural framework that is required to enable scalability, performance, availability and security for cloud services. It is only a matter of time before we see the IT industry recognizing the need to move beyond server virtualization and incorporate virtualized network and storage resources3 to enable dynamic provisioning of resources end-to-end. At this point, the cloud IT industry would do well to adopt a telecommunications-style IN model4 and implement application FCAPS management and a Virtual Resource Mediation Layer (VRML) to enable a 800 Service Call Model that can provision CPU/memory, bandwidth and storage resources dynamically based on application requirements. Using this model, application resource optimization based on application workload needs and business constraints becomes as simple as making a phone call. Service creation, delivery and assurance will become very similar in reliability and performance to those offered by the Telecommunications IN Services platform. Current IT emerged from a server-centric architecture that later evolved into a client-server architecture to accommodate network-based computing. Optimization in these architectures centered primarily on server resources. With the shift to network-based services, a next generation network-centric mediation layer is required to optimize the services platform for massive scaling and interoperability. By providing the mediation between virtualized computing, networking and storage resources, the VRML will become a network Operating System (OS) and its domination by a single vendor can create monopoly that may not be in the best interest of cloud computing.
Pros and Cons of Cloud Computing:
In cloud computing models, customers do not own the infrastructure they are using; they basically rent it, or pay as they use it. The loss of control is seen as a negative, but it is generally out-weighed by several positives. One of the major selling points of cloud computing is lower costs. Companies will have lower technology-based capital expenditures, which should enable companies to focus their money on delivering the goods and services that they specialize in. There will be more device and location independence, enabling users to access systems no matter where they are located or what kind of device they are using. The sharing of costs and resources amongst so many users will also allow for efficiencies and cost savings around things like performance, load balancing, and even locations (locating data centers and infrastructure in areas with lower real estate costs, for example). Cloud computing is also thought to affect reliability and scalability in positive ways. One of the major topics in information technology today is data security. In a cloud infrastructure, security typically improves overall, although there are concerns about the loss of control over some sensitive data. Finally, cloud computing results in improved resource utilization, which is good for the sustainability movement (i.e. green technology or clean technology.)
CONCLUSION:
In this paper, current trend in cloud computing have been analyzed and compared with the evolution of the telecommunications Intelligent Network (IN). A new reference model for the next generation datacenters that will enable both public and private clouds to be massively scalable and interoperable has been proposed.
Figure 4. Next Generation Cloud Infrastructure with VRML
Learning from the lessons of the past, the paper proposes a next generation Virtualization deviation Layer that goes beyond current server virtualization and integrates network and storage virtualization to enable seamlessly unified management. The VRML layer allows the creation of next generation virtualized computing, network and storage devices using “dumb” COTS components, while integrating into current generation architectures with plug-in adapters. This will allow gradual migration5 from current generation applications to SCN, providing the next generation services architecture in massively scalable and globally interoperable cloud platforms. The proposed platform can help transform IT infrastructure to bring it on par with telecommunications and Internet platforms that can scale massively while delivering reliable, and optimal performance along with fine grain security controls. Our envision transforming the datacenter into a “central office” for enabling application connection, (in this case, the connections of multiple applications with computing, 5 Migration of a large base of current applications to the cloud without interrupting the services they are currently providing will be an essential requirement for the SCN infrastructure. Virtualization and resulting dynamic provisioning capabilities and the 800 service call model will allow SCN to build cloud based mediation applications that interface with current generation storage systems through their management systems. Similar approaches have been adopted in the past in migrating legacy telecommunications system to IN platforms using plug-in adopters and mediation and conversion functions. network and storage resources) much like the telecommunications “central office” connects billions of people anywhere in the world and assures those connections even in the case of disasters. Developing the proposed VRML platform will require implementing a new distributed computing model, the 800 service call model, dynamic end-to-end FCAPS management, and signaling for business priority based resource allocation - all borrowed heavily from the telecommunications domain.
The paper also proposes that to be successful, VRML must be defined through a standards based RFI process with leadership driven by global standards bodies such as the IETF or ITU. The evolution of the telecommunications network and the Internet has demonstrated the success of this approach. While ITU provided top down standards development, IETF followed bottom up request for comment RFC process. For clouds to be massively scalable and for both public and private clouds to become globally interoperable, the role of the VRML is critical and must be vendor agnostic. We believe that the next generation cloud evolution is a fundamental transformation – and not just an evolutionary stack of XaaS implementations, which will enable global service collaboration networks utilizing optimally distributed and managed computing, network and storage resources driven in real-time by business priorities.
No comments:
Post a Comment