Clearing the Clouds Understanding cloud computing Ali Khajeh-Hosseini ST ANDREWS CLOUD COMPUTING CO-LABORATORY Cloud computing There are many definitions and they all differ Simply put, cloud computing is a model for delivering IT as a Service. Clouds refer to the actual data centres that house the hardware and software US National Institute of Standards and Technology working definition: Cloud computing has Five characteristics Three service models Four deployment models Clearing the clouds, November 2009 2
Characteristics 1. On demand self service: you can start using computing resources at anytime without needing human interaction with cloud service providers. Computing resources can be storage, processing, memory, network bandwidth, VMs... 2. Broad network access: you can access those resources over the network using laptops, mobiles phones etc. 3. Resource pooling: the computing resources are shared by multiple users (multi tenancy) 4. Rapid elasticity: you can scale up or down the amount of resources that you re using very quickly 5. Measured service: resource usage is metered by measuring your storage, CPU hours, bandwidth usage etc. Clearing the clouds, November 2009 3 Service Models Infrastructure as a Service Low level of abstraction, most flexible, dealing with virtual machines Platform as a Service High level of abstraction, less flexibility, dealing with your application code and your provider s APIs Software as a Service Using software that others have developed and offer as a service over the web Clearing the clouds, November 2009 4
Deployment Models Private cloud: controlled and used by one organization Community cloud: used by several organizations Public cloud: available to the general public Hybrid cloud: mixture of the above, allows cloud bursting Clearing the clouds, November 2009 5 StACC St Andrews Cloud Computing Co laboratory launched in April 09 Local investment of about 0.5 million covering PhD Studentships Experimental hardware platform Technical support Our aim was to explore the research potential of CC bringing together researchers in distributed systems, high performance computing and systems engineering Clearing the clouds, November 2009 6
StACC Experimental Cloud The StACC private cloud is now operational 10 servers (Cloud controller, storage server, 8 * 4 core general purpose servers) Running Eucalyptus open source software Objectives To understand what s involved in setting up and running a cloud (more than you might think!) To provide an experimental platform that we can measure To provide us with a platform that we can extend (and break) Clearing the clouds, November 2009 7 Cloud Service Consumers Software as a Service Platform as a Service Infrastructure as a Service Is it cheaper? Is it secure? How will it effect my work? Private Clouds Public Clouds Hybrid Clouds Community Clouds We did an extensive literature survey to see if we can find out the answers... Clearing the clouds, November 2009 8
Costs Should I lease or buy? Walker modelled the cost of a CPU hour when purchased as part of a server cluster and compared it with Amazon EC2 2 scenarios, purchasing: a 60,000 core HPC cluster a compute blade rack consisting of 176 cores Model showed that it s cheaper to buy in both scenarios assuming that CPU utilization is very high and electricity is cheap Good first step but far too narrow in scope, what about costs of housing the infrastructure, installation and maintenance, staff, storage and networking? E. Walker, The Real Cost of a CPU Hour, 2009 Clearing the clouds, November 2009 9 Costs Deelman et al. used simulation to calculate the cost of running a data intensive astronomy application on Amazon s cloud Highlighted the potentials of using cloud computing as a costeffective deployment option for data intensive scientific application Assumed the cost of running instances on AWS EC2 are calculated on a dollar per CPU second basis, i.e. they normalised the costs But, AWS charge on a dollar per CPU hour basis and charge for a full hour even for partial hours. So launching 100 instances for 5 minutes would cost 100 CPU hours Makes a significant difference in costs E. Deelman, G. Singh, M. Livny, B. Berriman, J. Good, The cost of doing science on the cloud: the Montage example, 2008 Clearing the clouds, November 2009 10
Costs Kondo et al. investigated the costs of using cloud computing for desktop grid projects such as SETI@Home They found that deploying the servers used for the SETI@Home project on Amazon s cloud would cost 40% less than using their university s data centre Didn t include upfront server purchasing costs or staff costs Cloud computing would look even more attractive if they did D. Kondo, B. Javadi, P. Malecot, F. Cappello, D. P. Anderson, Cost benefit analysis of Cloud Computing versus desktop grids, 2009 Clearing the clouds, November 2009 11 Costs So is it really cheaper? Not really sure, it all depends on the specific scenario and what you include in your calculations We re developing a tool to help users decide for themselves... Clearing the clouds, November 2009 12
Security Security concerns are often mentioned in cloud computing but not much research has been done to address them The Cloud Security Alliance has published a set of security guidelines in the form of problem statements and issues that need to be considered by users Most concerns are about loss of control over physical hardware which then lead on to legal issues... Clearing the clouds, November 2009 13 Legal Issues Most legal issues are related to the cloud s physical location, which determine its jurisdiction Amazon have data centres in 2 regions (US and Europe) so they can deal with these issues But the cloud s nature means that users don t know (or care) about this information: it s all in the cloud Location is important because cloud computing increases the control of governments and corporations over resources*. Cloud computing brings together vast amounts of data and computing resources in centralised data centres, compared to how they are currently hosted in geographically dispersed locations It s unlikely that these jurisdiction issues will stop the use of cloud services * P. T. Jaeger, J. Lin, J. M. Grimes, S. N. Simmons, Where is the cloud? Geography, economics, environment, and jurisdiction in cloud computing, 2009 Clearing the clouds, November 2009 14
Effects on Work How will cloud computing effect the work of IT departments? Their authority has been diminishing over the last few decades, from mainframes to PCs Cloud computing is going to decrease their authority further Users are turning into choosers * who can replace the services provided by the IT department with service offered in the cloud * R. Yanosky, From Users to Choosers: The Cloud and the Changing Shape of Enterprise Authority, 2008 Clearing the clouds, November 2009 15 Case Study To study these issues, we performed a case study examining the relative costs of migrating from a locally provided host to a host provided on Amazon The system studied was a fairly small data acquisition IT system from the Oil & Gas industry. The system had a contract value of 104,000 and was deployed in a local data centre Our focus was socio technical what are the human and organisational issues that influence such decisions Clearing the clouds, November 2009 16
Case Study Company A (major oil company) Minutely data Offshore oil rig Tape Drive Database Server Application Server Firewall Company A Offshore oil rig Company B (IT solutions company) Company B Amazon s Cloud Company C (end users) Company C Clearing the clouds, November 2009 17 Case Study Infrastructure costs over 5 years In house Cloud 57,000 25,000 Around 55% cheaper Support Issue 218 support calls over 5 years 20% of them would be eliminated Backup, power and network issues would be Amazon s responsibility Clearing the clouds, November 2009 18
Interviews We took these findings and presented them to various people in the company and interviewed them Technical manager: It looks attractive We would lose leverage over support contracts Concerned about security and protection of virtual machines Some corporations veto data going over port 80 Support staff: Very negative about cloud computing Feel threatened Don t want to lose control over hardware Clearing the clouds, November 2009 19 Interviews Project manager: Hard to sell this idea, there s no data centre to show clients Difficult to alter corporate security policies Easier to manage cash flow Reduces required skill set and overheads Business development manager: We are under pressure to reduce electricity usage It presents new opportunities for us, e.g. running training courses in the cloud Clearing the clouds, November 2009 20
Decision Support System Cloud computing has sparked a huge interest in the IT industry Many organizations are thinking about migrating their systems to the cloud. During this period, many migration decisions need to be made, what to keep in house, what to migrate These aren t simple decisions and there are a range of issues that need to be considered when making these decisions: economic, technical, organizational, risks etc. We are developing a decision support system to help people make those decisions Clearing the clouds, November 2009 21 Summary Cloud computing is still in its early days We are currently at the start of a transition period, the shift towards cloud computing could take many years Not much research has been done about issues effecting cloud service consumers Clearing the clouds, November 2009 22