Since 1993 we have maintained a stereo projection facility to create an interactive visual environment for computational molecular modelling. A projected computer screen serves as a window to three dimensional images which are easily viewed by groups of up to eighteen people. In addition, a haptic input device allows for users to actually feel the forces as they are applied to the molecular systems. More information on the facility is available here.

Over the past several years we have held several tutorials to help other research groups design and build their own PC compute clusters. These tutorials have consisted of a lecture explaining how to specify, design, and purchase a cluster to your local needs and facilities, followed by a hands-on session where users actually build a cluster using commonly available clustering software and our own hardware. More information is available here.

Since 1993 the Resource has been using compute clusters to cost-effectively perform Molecular Dynamics simulations. We built our first Linux PC cluster in 1998, and have continued to configure newer, faster, and larger systems approximately every two years since. Currently we maintain three clusters of 24 dual-processor Athlons, each capable of simulating a 100,000 atom system at 36hrs/ns. The total cost for this setup was less than $100,000. More information is available here.

In a research environement, data integrity is at least as important as creating the data in the first place. To ensure that files are not lost to user or hardware error, we have developed an in-house backup solution, based on Sun's ufsdump and ufsrestore commands. Files are dumped nightly to disk, and monthly to tape; several copies are kept of each. These dumps allow us to easily restore lost user files, as well as recover from catastrophic disk failures. Additionally, we perform twice-annual full backups of the file system using GNU tar, so that the data can be read back from any system ten years from now.

Servers that store data and serve content to the outside world must be as stable as possible. As such, we rely on three classes of stable servers to store our data and share our work. Four SunFire 280Rs store and share our data throughout the Resource's machines and beyond, acting as file, web, and mail servers; five Sun ES250s offer additional file storage capacity with years of uptime; and five SunFire V100s/Netra X1s offer additional services, such as name service, printing, SMB, and external logins. The full list is available here.

As the Resource simulates larger and larger molecules for increasingly long times, the need for disk space has grown exponentially. Our local network currently hosts 20 terabytes of hard drives, divided across seven file servers. In addition to a five gigabyte disk quota for their home directories, users can store up to three hundred gigabytes of regularly backed-up data in a shared Projects space. Each user also receives a 120 gigabyte local disk on their desktop workstations, for storage of local data that doesn't need to be backed up or is already stored elsewhere. An additional half-terabyte of shared space available as scratch space for all users. More information on our disk space partitioning is available here.

While many research groups choose to work behind a firewall, the Resource has chosen to stay open to the world and instead focus on making sure each system on the network is well-defended. Generally, the only service open on a given system is SSH, for remote access. User passwords never go over the network in the clear, where they could be intercepted; instead, all relevant connections are encrypted using SSL or similar technologies (IMAP/S, SSH, SFTP, etc), and the only unencrypted traffic is web, mail, and other unauthenticated traffic (and even these protocols offer SSL capabilities). Systems and services which cannot be fully secured are TCP wrapped, to remove access from unapproved systems. New systems are scanned to ensure that no unauthorized ports are open. Our system administration team generally patches holes in our open services within six hours of their discovery. Windows systems are patched nightly with SUS.

In order to properly analyze our data, each user must have a powerful graphics workstation on their desktop. These desktops are primarily Athlon 1333 workstations that used to serve as cluster machines; each one has at least 768MB of graphics and an nVidia GeForce2 GTS or GeForce4 Ti4600 graphics board, with a 120GB scratch drive for the local user. We also run several mid-range Sun visualization workstations (SunBlade 2000 systems with XVR1000 graphics) that offer better stability and the capacity to view molecules in 3D. Apple G4 systems are used for administrative work. The full list is available here.

When the power of our desktop workstations is not adequate to properly view and manipulate the molecules we study, we must use larger systems with additional memory, processing power, video, and input devices. Our primary visualization workstations are pair of dual-processor SunBlade 2000 systems with dual-XVR1000 video and four gigabytes of memory, equipped with stereo emitters, dual monitors, and SpaceBall input devices for the best user experience; one of these also runs our 3D projection facility. We also have an additional three single-processor SB2000 systems, a Sun Ultra 80, two Apple G4s, and a P4-1700 available for public visualization use. The full list is available here.

To properly develop our local software for our diverse user base, our developers need access to just as diverse a variety of hardware and operating systems, and we provide this as well as we are able. On the hardware side, in addition to our standard PC and Sun workstations we maintain development systems from Hewlett-Packard, IBM, SGI, Microway, and Apple. The full list is available here.