We now have a wiki which contains more details and updates.
The Yale CS Cloud is a private heterogeneous cloud for research in Computer Science. It allows groups within the department the ability to share advanced computational resources without needed to fund the entire costs on their own. Unlike other clusters on campus, the CS cloud allows for lower level access to hardware and software, allowing for the flexibility to experiment.
The cloud runs on Open Stack with a Ceph storage backend. Basic provisioning of resources can be done through the Open Stack Dashboard and through the API. More advanced customisation is possible upon request.
Compute (Zone 1)
120 physical cores divided over 5 nodes. Each node has 6 120GB SSDs and 3 AMD FirePro W8100 8GB GPUs. This zone has a one to one correspondents between virtual and physical resources, limiting the possibility of interference.
Compute (Zone 2)
A heterogeneous collection of hosts that support a large number of long term VMs. Like a traditional cloud, Zone 2 is over-provisioned, making it perfect for running traditional services.
Compute (Zone 3)
A collection of bare-metal nodes that can be automatically provisioned through the dashboard. Once provisioned, users have full control of the hardware and software.
192TB of physical storage divided over 4 storage nodes. Each node also has a 400GB high speed SSD cache.
Zone 1 and storage is connected by redundant 56Gbit Ethernet connections. The rest of the nodes are connected by a combination of 10Gbit and 1Gbit Ethernet.
Participating Groups and Labs
How will the new cloud infrastructure effect existing servers?
Depends on the server. Some servers will be left to function as they currently do. A few will be formatted, and included as "bare-metal nodes" nodes. This means that they will be allocated through openstack. However, once they are allocated, the user will have full control of the underling hardware of the node. (i.e. there will be no VM between the OS and the hardware). Finally, a number will be converted into hosts for VMs.
The current plan is below. However, it subject to change. We plan to keep korz as is. The HP blades, galorz and morz will be converted to hosts for VM. Finally, smorz and the dell blads would be allocated as bare-metal nodes. The rest of the servers are allocated for specific tasks. Their currently no plans to use them.
Will the storage servers be backed up?
We do not have the funding to back up the entire storage server. We recommend that you assume that any data stored on the storage servers is subject to deletion at any time. Please regularly back up any critical data you may have.
How is access to the cloud prioritised?
Groups that have helped fund the cloud will have property over other groups within the department. When multiple groups need the same resource, we will try to work with each group to schedule usage in a manner that is fair and takes each groups deadlines into account.