. Alice is a PC cluster, a pile of PCs, a Beowulf. It consist of 25 computer boxes connected together with an Ethernet. The head or the front-end has one AMD Athlon K7 running at a clock frequency of 700 MHz mounted on a ASUS K7V motherboard. The processor has 256 Mbyte of primary memory. It has also 120 Gbyte of ATA disk and a GeForce DDR graphics board. For communication, a Gigabit Ethernet NIC (Netgear GA620) is connected to the cluster switch and there is also an Intel 100/PRO+ NIC installed for communication with the rest of the world.

The other 24 nodes has the same amount of processors and memory. The secondary memory consist of one 6.4 Gbyte ATA disk and a FastEthernet NIC provides the network connectivity to other nodes in the cluster. An ATI Rage graphics card is installed. Its main purpose is to provide initial access to the BIOS settings. The nodes as well as the front-ent are assembled and delivered by Advanced Computer Technology AB.

Only the front-end has connection with the rest of the world. The other nodes are connected to a HP ProCurve 2424M (24 ports) FastEthernet switch. The switch has been extended with a Gigabit interface to connect with the front-end.

Most of the software on Alice is open source.

Operating System

The operating system on Alice is Linux. RedHat is the used distribution and the current version is 6.2. The Linux kernel is upgraded to 2.2.16.

Programming Environment

The following compiler suites are available on Alice:

• EGCS 1.1.2 (C, C++, and F77)
• PGCC 2.95-2 (C, C++, and F77)

Communication APIs

To easily utilize all boxes in a cluster, a efficient communication API is needed. The most well known APIs that use the message passing paradigm in parallel computations are MPI and PVM. Both are available on Alice.

Two MPI implementations are installed and configured:

• LAM from University of Notre Dame and
• MPICH from Argonne National Laboratory.

For PVM, the well known distribution from Oak Ridge National Laboratory is used.

Syncronization

To simplify installation of the same image on all machines, VA SystemImager has been used. The same software together with rsync and a couple of home-brewed scripts are used to keep the nodes in sync.

 TIME      WALL
 N         26000
 NB        60
 P         4
 Q         6
 LU Time   1158.95
 Sol Time  2.88
 MFLOPS    10086.12

         CLASS W        CLASS A        CLASS B       CLASS C
         1 proc         25 proc        25 proc       25 proc

 MG                      725.61          780.46       912.23
 LU                     1362.34         1430.34      1406.89
 IS                        6.70            7.10         7.48
 EP                       44.07           44.08        44.18
 CG                      190.15          185.30       212.13
 SP                      841.93          826.22       969.12
 BT                     1785.73         1703.49      1992.58

 Function  Rate (MB/s) RMS time Min time Max time
 Copy:     447.4838     0.0717   0.0715   0.0724
 Scale:    419.8648     0.0762   0.0762   0.0764
 Add:      562.4103     0.0863   0.0853   0.0865
 Triad:    473.2887     0.1015   0.1014   0.1015

All the available time on Alice is dedicated for use by The SKF Group. SKF are mainly using Alice for running a program named BEAST, (Bearing Simulation Toolbox). It simulates the dynamics of a rolling bearing by solving the general differential equations of motion for all components.

If you are interested in testing your application on Alice and see what kind of performance and scalability is possible to get on this architecture using your own software, please contact support.