Intelligent System Supervision - Cloud Computing

Anmerkungen:

• Ubiquitous / network access o Not only the Internet  On-demand self-service  „Resource pooling” o „Multi-tenant model”: the same resource pool is used to serve many tenants o Dynamic allocation of resources & services to tenants o Tenant control of that: rather/very/extremely limited 
•   Elastic scale-out and scale-in o Seemingly infinite amount of resources, o that can be rented (and released) at any time    Metered/utility-like services o Some notion of „service usage” metric („how much service is used”) is introduced o Payment is at least partially based on „actual usage” • N.B. modulo time granularity (hour, day, month...) 

Anmerkungen:

•  Capability: using the applications of the provider o Access: usually thin client (browser!) o In all ernesty, not a new concept  Examples o Google Apps o Salesforce CRM o LotusLive o Microsoft Business Productivity Online Suite (BPOS)  Hugely successful in some domains: collaboration, accounting, CRM, ERP, HRM, CM, PM, ...

Anmerkungen:

•   Capability: deploying a custom application into a „rented” runtime environment (usually app. server container) o Predefined platform services o Predefined languages and APIs o The environment can be configured to some extent o The application model may be constrained • E.g. „no direct access to the file system”, „no direct access to OS networking”, ...  Google AppEngine  Microsoft Windows Azure Platform  Amazon Beanstalk 

Anmerkungen:

•  Capability: access to fundamental computational resources o Most fundamental manifestation: „renting virtual machines” o You run what you like on the OS you get o Resources are usually either logical or virtualized o Meaning no direct access to the hypervisor & HW level  Amazon Elastic Compute Cloud (EC2) o Xen-based virtualization o Full ecosystem o Note: as a rule, general purpose cloud service

1. Amazon Elastic Compute Cloud - EC2

   Anmerkungen:

   • For building something bigger: e.g. AWS CloudFormation. Amazon Elastic Compute Cloud is a central part of Amazon.com's cloud computing platform, Amazon Web Services. EC2 allows users to rent virtual computers on which to run their own computer applications.    

1. When don't use

   Anmerkungen:

   •   Constant load  Legacy systems  Security, legal obligations and regulatory compliance   
2. Demand and Capacity
3. Deployment models

   Anmerkungen:

   • Private: for a single organization with internal tenants. May be locally managed and owned by the organization.           Community: a cross-organizational community shares the cloud resources (can be closed) o Gorvernment, healthcare, finance, education, ...             Public: open (!= free (as in beer)) access. Operator owns the service infrastructure  Hybrid: composition of two or more clouds with data and application portability 
4. Cost optimization

   Anmerkungen:

   •  There are costs involved with... o Demand left unserved o Overcapacity  The unit price unit price of a cloud can be higher than investing in your own infrastructure, but...  ... the same goes for hiring a car or a hotel room!  Hybrid clouds: private – base load, public: varying o Nontrivial optimization problem o Still have to guess predict future demand 

1. Scaling resources

   Anmerkungen:

   • "Scale out” o Parallelizability? o "webscale” technologies o Clustering and replication o (Mongo DB Is Web Scale) 
   1. Capacity planning

      Anmerkungen:

      •   Too low capacity: unserved demand o E.g. planning for average load   Too high capacity: not economical (?) o E.g. planning for maximum load   Physical („in-house”) capacity is usually slow and costly to increase...o ... and decrease!   (One-time) investment and commitment   Demand? 
      • ...surges that can’t be predicted   The cloud is scalable, let's use it!    For larger businesses with existing internal data centers, well-managed virtualized infrastructure and efficient IT operations teams, IaaS for steady-state workloads is often no less expensive, and may be more expensive, than an internal private cloud.” 

1. Paralellizable loads

   Anmerkungen:

   • More and more embarrassingly parallel problems and „scale-out” applications  NYT TimesMachine [12]: public domain archive o Conversion for the web [13]: Hadoop, few hundred VMs, 36 hours  Due to usage based pricing costs approx. the one VM  In a sense, „speedup for free”