- 2 Air-conditioned, 2-meter high server cabinets
- Computing Nodes:
Cumulatively, there are 13 computing nodes with the following specifications: - 1 Rack node, with 96 logical cores (48 physical cores @ 3 GHz, Dell R740xD)
- 2 Rack nodes, each with 72 logical cores (36 physical cores @ 3 GHz, Dell R740)
- 5 Rack nodes, each with 56 logical cores (28 physical cores at 2.6 GHz, Dell R730)
- 5 Tower nodes, each with 12-48 logical cores (6-24 physical cores at 2.67-3.7 GHz)
- Processing Memory:
Each rack node is configured with a large amount of fast memory (>500 GB) to support heavy-duty processing operations: - Dell R730 Rack Nodes: 640 GB of memory each
- Dell R740 Rack Nodes: 640 GB of memory each
- Dell R740xD Rack Node: 768 GB of memory
- Tower Nodes: 32-256 GB of memory each
- GPUs:
A total of 20 GPUs distributed across the nodes: - 1x NVIDIA Tesla V100 32GB, 5120 cores
- 5x GTX 2080Ti 11GB, 4352 cores
- 10x GTX 1080Ti 11GB, 3584 cores
- 1x Quadro K5000 4GB, 1536 cores
- 1x Quadro M2000 4GB, 768 cores
- 2x Quadro K4000 3GB, 768 cores
- 1x Quadro K2000 2GB, 384 cores
- 188 TB of network storage (170 TB available with RAID5)
- 10 Gbps network switches
- Turn your physical computer into a VM for an OpenStack Cloud
Yes, you got it right. You can create a VM out of your personal computer. This VM running in an OpenStack cloud (such as the RVL Cloud) will behave just like the physical computer you are currently using. NOTE: When one creates a VM out of a physical machine in this manner, typically one does not copy over one's home directory. But, if you want, you can include your home directory also in the VM. - Save a physical OpenStack node as a tarball
Let's say you want to add a new physical node to an OpenStack cloud. An efficient way to do that is to create a tarball from one of your current nodes and copy it over to the new node. This script shows how you can do that. - Install the tarball for an existing OpenStack physical node in a new physical node
After you have created a tarball from an existing OpenStack node, you can use this script to install the tarball in your new node.
Overview
The main reason for why we created the RVL Cloud is that practically every research project these days has some sort of a big-data and deep-learning component to it. A cloud based execution of these projects makes it much easier to customize the resources needed for solving such problems. Additionally, some of our research projects now require deliverables in the form of virtual machines that must run on a commercial cloud platform like the Amazon Web Services.
These factors have heavily influenced how computing is done in RVL at Purdue. Several of our research
projects involve large datasets of satellite imagery. These datasets can be terabytes in size and are
typically processed by GPU enabled VMs. Other examples include processing of video for tracking humans
and objects, detection of hazardous materials in airport security systems, retrieval of information from
large textual datasets and software libraries, automatic bug localization in large software systems, etc.
RVL Cloud runs on the OpenStack framework. This is the home page for our RVL Cloud where we hope to share
our experience and knowledge in setting up a powerful open-source cloud computing platform in a university
setting.
We are frequently asked as to why we created our own cloud platform. Why not just rent time on, say,
Amazon Web Services? To answer, an in-house cloud platform allows the students to experiment with the
cloud architecture itself and thus develop deeper insights into the workings of a cloud. Additionally,
the inter-node communication in an in-house cloud can be much faster than in a large-scale commercial
operation.
Hardware Specifications
Bash Resources
If you are interested in setting up your own OpenStack cloud, here are some bash script that you might find helpful. A common strategy in setting up a new cloud platform is to first install a cloud controller in one of the physical nodes. Subsequently, you create your first general-purpose compute node for the cloud on another machine. Finally, you duplicate the compute node in the other machines you are deploying for the cloud. In the cloud platform thus created, you can then start creating virtual machines for whatever application you have in mind. The scripts that are listed below should help you with all these phases of an OpenStack cloud set-up.