UKUUG 2011 – OpenStack

Print Friendly, PDF & Email

I got a t-shirt! Whoop! I’m sold! I’m such a t-shirt whore. ;-)

But seriously, next to Ceph, this would be the next take-back-to-work talk. OpenStack is probably the first Virtualisation cluster management suite built on open standards for open platforms such as Ubuntu making it very accessible. OpenStack is the amalgamation of the storage component of Rackspace’s internal datacentres, call Swift and the forked project by Nasa of some of Eucalyptus clone of Amazon Web Services EC2 called Nova. It’s goals are simplicity in design for easy adoption and massively scalable. It’s heavily community based with all access to development platforms placed on community services such as Launchpad using bazzar for collaborative code development.

Swift storage works on a object storage service ODS model similar to other cluster suite filesystems like Hadoopfs, Lustre and Ceph. Data address “rings” act as a proxy to store location information which can lookup container and account and runs on standard commodity hardware (no RAID).

Nova is based on the AWS API but also supports Rackspace own API for automating VM creation. The API connects to a group of services plumbed together by a message queuing protocol service (currently RabbitMQ). Separate services which can be separated out into different cluster machines are a scheduler, volume manager, network manager, compute node and an external data store for cluster configuration data. This last component is seen as unneeded and only transitional at this stage of development. The model is designed to be modular and based on multiple open standards software; e.g. API = WSGI, computation node = Xen, libvirt (e.g. KVM), Hyper-V, volume manager = LVM, iSCSI, sheepdog, scheduler = pluggable (chance), etc…

Should be easy to test out an installation with the next release of Ubuntu Natty Narwhal will include both Openstack and Eucalyptus. Which is nice…