|
Laboratory of Molecular Biophysics
|
Next: Crystallisation Facility, Up: Facilities, Return to: Contents.
The main laboratory computing facility has received only minor changes over the last year. It continues to be based on the following elements:-
In February 2001, Mark Sansom's group replaced their 10-processor Silicon Graphics Origin 2000 with a 64-processor Beowulf cluster, supplied by VALinux Systems, following on from their successful experience with an 8-node cluster. The hardware consists of 32 nodes, each with two 750Mhz Pentiun III processors and 512MB of memory and a 15GB local disk, linked by a 100 Mbit/sec switched Ethernet, and via a routing node to the laboratory network. Each processor node is in a 2U rack unit, so the entire system fits in two large racks in the main computer room. Each node runs a local copy of the operating system (Linux v2.2.19). Cluster management is performed through a single 'mayor' node, which can provide remote access to each node at the console level, and monitor each node through the VACM management hardware.
The primary applications run on this cluster are Gromacs 3 for Molecular Dynamics calculations and Gaussian 98 for ab-initio Quantum Mechanics calculations. Each job can be run in parallel (LAM MPI library) on a selected number of nodes, controlled by the PBS queuing system.The continuing increase in performance, both in terms of CPU and graphics speed, of PC and Macintosh hardware has now virtually eliminated the traditional workstation from the market. Around 40 PCs (mostly Compaq AP200 or AP250) with Pentium III processors and high-performance graphics and 20 Macintosh G3 and G4 systems are now installed around the laboratory. Most of the PCs are running WNT, and using the X-windows package Exceed for interactive access to the central systems. However, a number are being run with the Linux operating system, allowing many packages, such as CCP4, to be run locally. The lastest Macintosh operating system, OS-X, is also a significant change from its predecessors, being based on an underlying Unix operating system with a Macintosh interface layered on top. This allows both Unix and traditional Macintosh applications to be run on the same system.
A number of Compaq Alpha and Silicon Graphics workstations are still available, and also a few X-Windows terminals that provide a useful supplement to the interactive devices available.
The highest-availabity disk storage continues to be the two Compaq Storageworks RAID arrays, which are dual-hosted by the main VMS cluster nodes. These arrays are now four and five years old respectively, and have lower transfer speeds and maximum disk sizes than current systems. Therefore, in response to the need for more disk storage, especially for for the trajectory files calculated in molecular dynamics runs and for image data storage, we have added two economical RAID arrays, each with a capacity of 525 GB, thus nearly tripling the available central storage. The units are G-Force RI arrays, each containing a RAID controller, eight 75GB EIDE disks which are configured in a RAID level 5 set for reliability, and an UltraSCSI host interface. By using IDE interfaces for the local disk communications, considerably economy is achieved. Each array can be hardware partitioned into a maximum of 8 partitions, but we have also used software partitioning to organise most of the space into 40GB partitions, each allocated to a specific user. This also conveniently matches the nominal cartridge capacity of our backup system, as the administration is simplified if a full backup of a disk partition fits on a single cartridge.
Although this provides a perfectly satifactory RAID array, there are a number of disadvantages compared with the Storageworks arrays:- changing partition sizes requires the whole array to be reinitialised; it does not support dual-hosting on a shared SCSI bus in the same way as the storage works arrays, and so for the time being is hosted solely by the AlphaServer 1200. The lack of a server failover ability is a serious disadvantage, as it means that the storage will we inaccessible if the server should fail or be shut down for maintainance. We shall approach this problem when considering the upgrade of the present system. In addition, there is a total of about 50GB of local storage on the AlphaServer 4100, also served to the other systems.
We continue to make regular backups of this storage, except for space specifically designated to scratch use or for 'image' storage, for which archival is the users' responsibility.
The Compaq Storageworks TL891 tape robot, installed last year, has given excellent performance and has been invaluable in providing sufficient capacity to continue our automated daily backups, despite the increase in disk space over the last year. Around 900GB of space is scheduled for backups, although a significant proportion of this is as yet unused. The administration of tapes has been improved by the installation of the the Media and Device Management System software (MDMS), which coordinates tape cartridge details, robot slot position, drive selection and cartridge movements. This also supports the HSM facility, and the shelved data has been migrated to higher-capacity tapes in the TL891.
|
The laboratory network continues to be based on a switched 100Mbit/sec Fast Ethernet network (Figure 1). This network has been extended to a few areas, such as the new electron microscope room and workshop in the basement, that previously had only an ad hoc connection. Around 120 computers (counting the Beowulf cluster as 1!) and printers are now connected to this network. The older FDDI, thinwire ethernet and Apple LocalTalk networks are still in place, but as we continue to replace older systems they are becoming increasingly redundant and will be phased out over the next year.
External connectivity has been improved by the upgrade of the University Janet connection to SuperJanet IV in March 2001, with a connection speed of 622 Mbit/sec, which we continue to access via our 100Mbit/sec link to the University Gigabit backbone.
Network security is a continuing issue, with ever-more sophisticated attacks taking place. The University security team control some access at the University Firewall, but of course there is a limit to the controls they can apply to the University network as a whole. We have also continued maintain the security of our systems, by upgrading operating systems or applying patches as they are released, and in addition disabled many non-essential network services, especially on workstations.
Next: Crystallisation Facility, Up: Facilities, Return to: Contents.