![]() |
Project Background: In modern chemistry research, computational resources are a very important tool. While such calculations are not always extremely accurate, they are still very useful. A wide range of levels of theory are available, and represent a range of accuracy. Some calculations are fast, but rough. Others take a long time ("computationally expensive"), but are much more accurate. In fact, the amount of time a calculation takes to finish tends to increase exponentially with its level of accuracy. The size of the molecule under consideration also important. Some extremely simple systems have the potential to finish in just a matter of minutes, but when one tries to run such a calculation on a much more complex system such as a protein, the calculation may take many years to finish. A wide range of software has been written to perform a large variety of calculation types, from the simple calculation of energy or entropy to the visualization of molecular orbitals and electrostatic potential maps. While computational and model chemistry is a magical tool that has seemingly infinite use in nearly all fields of chemistry, it comes at a price: CPU Power. Effective model chemistry research requires a lot of CPU power that costs a lot of time and money. Learning how to set up an effective means to do computational work can also be a barrier to efficient model chemistry research. This often requires construction of a computer cluster - a group of many computers working together. A larger computer cluster, however, brings increasing levels of complication. A fascination with computers, an interest in chemistry, and the desire to learn about things that fit into one or both of these categories all motivated me to begin this project. Gustavus owns a computational chemistry cluster, however in recent months it has developed a number of problems and inefficiencies; it needs some maintenance. I made it my goal this month of January to work on rebuilding Bert and Ernie: The Gustavus Computational Chemistry Cluster in the basement of Nobel Hall. | ||||||||||||
| The Current
Setup:
| |||||||||||||
| Current Cluster
Inefficiencies: Limited number of nodes ... We need to get some more nodes running. Hardware maintenance is needed on a number of nodes ... Broken hard drives, insufficient memory, worn out fans, and a number of other such plagues are keeping many of our nodes from operating. User accounts must be created on each node individually ... In order to create a new account for someone, it is currently necessary to edit each individual machine's user account database. Glitches in NFS cause interruptions of service ... On rare occasion, the NFS shares that are necessary to run the Gaussian software tend to drop, sometimes causing jobs to fail. Outdated Software ... Makes the systems vulnerable to hackers (as we learned last winter break)...
| |||||||||||||
| The New (Planned)
Setup:
| |||||||||||||
| Points About The
New Setup: The nodes will be on a private network ... This will make many things a little bit easier, including getting the systems registered on the gustavus network every year. Also, it will increase security just a little and make the NFS mounting a little more reliable. Internet Access is routed to the private network ... They can receive security updates and/or other software if necessary. OpenLDAP will be used as a network authentication scheme ... Eliminating the need to add users separately to each node! Yay! Managing users will be so much easier.
| |||||||||||||
![]() |
Tasks To
Accomplish... 1) FIX THE BROKEN PARTS!!!
2) Install New Operating Systems!! - Fedora Core 8
3) Configure network interfaces on bert 4) Install and configure
DHCP (Dynamic Host Configuration Protocol) and DNS (Domain Name
System) 5) Enable IP
Forwarding
*nat ... YAY!!! The nodes now have access to the internet!!!! | ||||||||||||
![]() |
Next Steps... OpenLDAP (Network Authentication) Torque/Maui (PBS Batch Queueing System) Gaussian 03 and AMBER WebMO and WebAMBER |