OrangeGrid is the Syracuse University high-throughput computing (HTC) cluster. It is comprised of over 70,000 cores and several dozen terabytes of memory. The computers in the grid are optimized to perform a large number of parallel jobs, providing high processing capacity over long periods of time. The grid utilizes a mixture of dedicated nodes (60,000 cores) and scavenged nodes (10,000 cores).
HTCondor, developed with support from the National Science Foundation, manages the grid’s workload. Scavenged worker nodes are managed by HTCondor Virtual Machine Coordinator (CVMC), an application developed by SU’s Information Technology and Services department. These nodes are added to the grid by detecting when a desktop computer is idle, launching CVMC, deploying a custom virtual machine, and connecting it to HTCondor to receive work. The use of virtualization acts as a barrier that separates the researcher and their content from the user’s information on the same computer.
OrangeGrid is supported by ITS, Syracuse’s central IT group, and offered to researchers without cost. The grid is unique in that approximately 20% of the nodes on the network are upgraded annually as part of regular campus desktop replacement cycles, providing a notable increase in processing and memory capacity each year.
In 2015, Syracuse partnered with the Open Science Grid (OSG) to contribute OrangeGrid compute resources to researchers within the OSG framework. Additionally, cycles not consumed by SU or OSG researchers are put to use in the LIGO Einstein@Home distributed computing project.
Syracuse University’s Information Technology and Services (ITS)
OrangeGrid utilizes:
- Virtualization via Oracle’s VirtualBox
- Scheduling via the HTCondor HTC System
- Desktop components management via ITS’ HTCondor Virtual Machine Coordinator (CVMC)
HTCondor Virtual Machine Coordinator (CVMC) is a small application developed by SU’s Information Technology and Services (ITS) team to manage the multiple desktop components.
These components are distributed to desktop clients via Microsoft’s Active Directory. HTCondor, developed with support from the National Science Foundation, manages the grid’s workload. The computer’s task scheduler detects when its host computer is idle, starts up CVMC, and connects to HTCondor to receive work. When user activity is detected on the computer, research operations are immediately stopped. The use of virtualization acts as a barrier which separates the researcher and their content from the user’s information on the same computer.
Header Image Credit:
Barrett Lyon / The Opte Project
Visualization of the routing paths of the Internet.