Operations Director R&D Manager Flexiant Ltd Flexiant Ltd tsharif@flexiant.com gihan@flexiant.com What is Cloud Computing? Is it a software (SaaS)? Salesforce, google docs etc Is it a platform (PaaS)? Google App Engine Is it an infrastructure (IaaS) ? FlexiScale, EC2 Wikipedia definition- Cloud computing describes systems that provide computation, software, and data access services without requiring end-user knowledge of or dependence on the system's physical location and configuration. Infrastructure as a Service There are more aspects than PaaS or SaaS model. Virtualization holds the key. But virtualization itself is not enough! Things to consider: Storage Network Security Accounting Scalability Clouds vs Traditional Hosting Instant provisioning. Servers are now just a mouse click/api call away. No hardware maintenance. On demand resources. Adding extra RAM/CPU to your server is a few mouse clicks. Pay as you play model. No long term binding contracts. Throw away the servers, no more Capex. Ideal for development environments. Demo Building your own Clusters of physical nodes with considarable amount of RAM/CPU. Your choice of Hypervisor . Open source Hypervisors give more flexibility when building a cloud platform. Important consideration- The Hypervisor is just another component in the cloud stack. Storage Network. ISCSI, NFS & FC, always ensure an API/CLI is available. Using local storage will loose the ability to live migrate and live recover as well as recover data from a local hardware failure. Building your own Programmable Network. Linux routers are ideal. Switches that support multiple static VLAN . Important : Some coding to glue everything together. Keep it simple. Considerations Management of Servers. Ability to create, delete, start and stop multiple VMs simultaneously. Finding the ideal physical node to run a VM. Automatic recovery on physical node failures . Single point of failures. SAN, Routers and Switches. Minimum human intervention on day to day running of the cluster ensures a cost rffrctive solution. You cannot stop failures, but you need to recover fast Our Approach The idea of programming the data centre. Separate management services. Management of Nodes/Hypervisors. Management of Network . Management of Storage. A cluster manager will combine all management services. Business logic has to be built on top of the cluster management layer. Each layer has its own SOAP API. Distributing the working intelligence. Avoiding Dependencies Imagine Nodes are just containers. You should be able to move between containers . Nodes are PXE booted, rebooting will put it back to clean state. Be Hypervisor-Independent! Cloud based services should not worry about underlying fabric. Multiple SAN support . Abstract layers. Simple driver architecture. Allow us to easily use multiple SANs/Routers and Hypervisor s in the same management stack. What is in the Pipeline IPv6 Support. Ready for release. Affinity. Customer level control which would allow a customer to determine positive or negative affinity for VMs. Working in our lab environment. Elvira. Distributed block layer which can sit on top of any different vendor SANs. Question Time