A Review on "Above the Clouds: A Berkeley View of Cloud Computing (Armbrust, Fox, Griffith at.el.)" Introduction: Cloud computing is the physical form of "computing as a utility". What is Cloud Computing: Cloud computing refers to 1. application delivered as services over Internet 2. H/w and systems S/w in the data-centers that provides those services Services are referred as SaaS(S/w as a service). The Data-center hardware + software is called "Cloud" By: Rakesh Kumar Mallick ME., CSA Sr. No.: 4710-410-091-06888 Two types of cloud: 1. Public cloud (cloud is made available in "pay-as-you-go" manner to public, the service being sold is utility computing) 2. Private Cloud (internal data-centers of a business or an organization that are not made available to public) SaaS providers are Cloud users and they provide cloud computing facilities to end users. Top level can be recursive, in that smaller SaaS providers are users of bigger SaaS. End users can access the service "anytime, anywhere". From H/w point of view, aspects that are new to cloud computing: 1. Illusion of infinite computing resources available on demand, thereby eliminating the need for cloud computing users to buy extra resources immediately. 2. The elimination of up-front commitment by cloud users. 3. The ability to "pay for use of computing resources on a short-term basis as needed and release them as needed" Things that can be purchased in cloud: 1. Processing Speed 2. Storage space 3. Network Bandwidth to move data in and out of cloud over Internet Why to become a Cloud Computing Provider: 1. Make a lot of money by making a new cloud. 2. Getting more revenue by using under-utilized resources for cloud computing. 3. Establish a new business of cloud computing which will work well even if main business is not. Introduction of New Technologies because of cloud computing: Shift from "high-touch, high-margin, high-commitment" services to "low-touch, low commitment, low margin" services. Hence introduction to "pay-as-you-go" transaction services.
New Application Opportunities: 1. Mobile interactive applications (people want services in real-time. Queries and services that generally relies on large data set improves in speed when taken care by large data-centers) 2. Parallel batch Processing (If there is enough data parallelism in the application, users can take advantage of cloud, which is also cost effective; because cost of using 1000 servers for 1 Hr. is the same as using 1 server for 1000 Hrs.) 3. Rise of analytics (It is a special case of compute-intensive batch processing) 4. Extension of compute intensive desktop application (Some computation intensive desktop applications, simulations and mathematics software packages like matlab, mathematica are capable of using cloud computing to perform expensive computations) Classes of Utility computing: 1. Hardware virtual machine: Resources provided by cloud looks more like physical hardware which can be nearly entirely controlled and configured by users. eg. Amazon EC2. + : There is not a priori limit on the kinds of applications that can be hosted. The low level of virtualisation-raw CPU cycle, block device storage, IP level connectivity allow developers to code whatever they want. - : This make it difficult for Amazon to offer automatic scalability and fail-over. 2. Application framework: Application domain specific platform eg. Google AppEngine, Force.com, SalesForce business software development platform etc. + : Supporting traditional web applications, enforcing application structure of clean separation between a stateless computation tier and a stateful storage tier. - : Not suitable for general purpose computing 3. General Purpose Computing: Intermediate between complete application framework and hardware virtual machine. + : The system supports general purpose computing rather than a single category of application - : User cannot control the underlying operating system. Which one is better?? Different task will result in demand for different class of utility computing. So it depends on task. Highly managed cloud platforms can be hosted on top of less managed ones (recursive layers of SaaS Provider). eg. AppEngine can be hosted on top of Azure or EC2; Azure can be hosted on top of EC2. Cloud Computing Economics: Let's make some observations about cloud computing economic models: in deciding whether hosting a service in the cloud is benificial in long term. If: UserHours cloud *(revenue-cost cloud ) <= UserHours datacenters *(revenue-cost datacenters /utilisation) in making decisions about whether to move an existing service to the cloud. If (S/w complexity + cost of moving data from a legacy enterprise application into the cloud + Network transfer cost + computation cost) is less than not using cloud than moving to cloud is advisable Elasticity & Transfering Risk: Valuable to establish companies as well as startups. The cost per user hour without elasticity is high because of resources sitting idle most of the time. Same thing happens when demand is over-estimated and resources are purchased also.
We also don't want to underestimate the demand. Scale-up elasticity is not a cost optimization but it is an operational requirement. Scaling-up and Scaling-down both are flexible as per user request. Usage based pricing (pay-as-you-go payments) Comaring Cost: Should I move to cloud? (S/w complexity + cost of moving data from a legacy enterprise application into the cloud + Network transfer cost + computation cost + power, cooling and plant cost) <= cloud-cost Top 10 obstracles and Opportunities for Cloud Computing: Technical Obstracles to the Adoption of Cloud Computing: 1. Availability of service [Cloud may be down because of some technical errors] Use multiple cloud providers to avoid complete business down Use up-scale elasticity to defend against DDoS attacks 2. Data Lock-in [Customer cannot easily extract their data/programs from one site to run on another] Standardize the API so that a SaaS developer can deploy Services and data on multiple Cloud Computing providers. Enable Surge Computing i.e. Use of public cloud to do extra task of private cloud. 3. Data Cofidentiality and Auditability Deploy encryption, Firewell, VLAN. Accomodate National Laws via geographical data storage. Technical Obstracles to the Growth of Cloud Computing: 4. Data Transfer bottleneck Shipping of data to datacenters. Keep data in datacenterc for long time since it can be reused. Improve intra network bandwidth for clouds. It can be a bottleneck for inter data processing. 5. Performnce Unpredictibility I/O access and I/O intensive application might be slow. Use of flash memory. Improved Virtual machine support Gang-scheduling. 6. Scalable Storage illusion of infinite storage space 7. Bugs in Large Scale Distributed System Use better debugers. 8. Scaling Quickly Automatic scaling-up and scaling-down as per resource demand. Policy and Business Obstracles to the Adoption of Cloud Computing: 9. Reputation Fate Sharing [One customer's bad behaviour can affect the reputation of the cloud] Create reputation Guarding services (similar to trusted e-mail services) 10. Software Licencing pay-as-you-go licence Conclusion & Questions about the clouds of tomorrow: The long dreamed vision of computing as a utility is Cloud Computing. From cloud provider's view, they are getting extra revenue by availing their resources for Cloud Computing. From cloud user's view, they are getting benifited from cloud by getting extra resources at their critical times.
Cloud is really a boon for companie like Pixar etc., and for startup ones. Cloud's Future: Developers would design their next generation of systems to be deployed into cloud computing. Application Software will have sufficient parallalism to enjoy Cloud Computing. Hardware Design: Virtual machine concept will be in mind of designer during designing hardware. Qu. Left Behind: change in technology and prices over time virtualisation level ----------------------------------------------------------------------------------- Questions: 1. If avg. resource demand = x; peak resource demand = 10x. What should be the system configuration (should it be good enough for 2x demand or for 3x or something else. How to find the least/most configuration that someone should have for his e-business. Of-course people don't want to have plenty of resources which are most of the time under-utilized and neither they want to use cloud services most of the time). Where a company's system configuration should lie. example of some e-business: fire-cracker, greeting-cards 2. [This could be a wrong Qu.] Microsoft's Azure is suitable for general purpose computing (I think, it includes specific computing features like of Google AppEngines also). Why we are not shifting totally on Azure. Is Azure slower here than AppEngine. 3. Capacity mentioned under section 6.1 (also refer figure 2.c) is soft capacity. It depends on patience of customer (more patience high capacity). How to determine this capacity for different occations. 4. In section 6.1 introduction of "usage based pricing" is given. It means if a company hired some resources from cloud but not using it then it don't have to pay anything to cloud provider. Is it right for cloud providers in terms of revenue. If the above is correct then every e-business company must hire resources from a cloud provider (it does't matter whether their own resources are under utilized, because they don't have to pay anything for hiring but not using cloud resources). 5. Capacity decreases as load increases. Is it true?? 6. What are these terms: VLAN, Packet filter? 7. I equation 2 why cost datacenter is divided by utilization? 8. Reputation Fate Sharing? Things I like: 1. Paper is giving a good ideas about "How to select an economical good place to establish data-center" 2. Cloud Computing Economics model observations 3. Ideas to encounter DdoS attack. Things I don't like: 1. Paper seems to be too long. Things are repeated line by line at some places. It is like just filling more pages without giving extra info. 2. Third solution to overcome data-transfer bottleneck is given to reduce the cost of WAN by doing more research on WAN. (It is not at all a solution. It is a suggestion to bring down WAN cost. Researchers might fail to do so also). Suggestion: 1. There should be a statistical data on annual cloud provider's revenue and avg. annual cloud user's savings by purchasing resources from cloud instead purchasing it's own physical resources. 2. One more section could be added named "Why to become Cloud Computing Provider". This
part is presently under "What is cloud Computing" section. Future Work that I like to add: 1. Classes of utility computing seems to be not well defined (or I should say not complete). 2. There can be a reinforcement learning based method which can predict spikes (sudden and excessive use of any e-business site) size and capacity. This will be helpful in maximizing profit. 3. Spikes can be of two types: predictable & unpredictable [Spikes capacity is also having same types]. More work is needed to make decisions to overcome spikes problem.