The Janus supercomputer is a modular, massively parallel, and reconfigurable FPGA-based computing system for High Performance Scientific Computing. It takes the name from the ancient Roman god of gates and doors, of beginnings and endings, of passages and time, usually represented by two faces, one looking to the past, another one to future. After a gestation period of around two years, Janus was born in 2008 thanks to a very successful scientific collaboration between researchers from Italy and Spain. The institutions involved are: Università degli Studi di Roma “La Sapienza” and Università di Ferrara in Italy and the Institute of Biocomputation and Physics of Complex Systems (BIFI) –University of Zaragoza, Complutense University of Madrid and University of Extremadura in Spain. This scientific collaboration has been called The Janus Collaboration.

Both Italian and Spanish groups have a deep background in special purpose computers. For instance, the Italian group started to built in 1985 the APE (Array Processor Experiment) family computers dedicated to the study of Lattice Gauge Theories with very good results (see [1] and [2]). On the other side of the Mediterranean, in 1991 the group of Zaragoza designed and developed RTN (Reconfigurable Transputer Network) [5] dedicated to the U(1)-Higgs model (Lattice Field Theory) and Spin Glasses simulations. Later on, a second generation spin glass machine was completed in 2000 by the same group, called SUE (Spin Update Engine) [4]. From the innovative work by Pearson and Richardson [3] in the late 70’s or that of Ogielsky and Condon [4] in the 80’s to our days, profiting of the newest technological advances, several refined dedicated computers and improved clusters have been developed to study spin glasses.

The supercomputer Janus is composed by 16 boards. On each board, a bidimensional 4×4 grid of FPGA-processors is located and linked obeying periodic boundary conditions. Each of these processors is called SP (Simulation Processor) and carries on the simulations. A 17th FPGA is settled in the middle acting as a crossbar and called IOP (Inpu/Output Processor), in charge of all internal connections and external communications. All FPGA modules are Xilinx Virtex4-LX200.

 

The Janus Operating System (JOS) is compatible with any Linux-based platform. It comprises three parts:

  • JOSlib: libraries in Perl and C to control IOP devices),
  • josd: multiuser environment for resource abstraction and concurrent jobs management)
  • jlib: set of SP firmware modules for scientific applications and C libraries to control them via the josd.

 All SP’s in Janus are reprogramed using the VHDL language.

 For a more detailed explanation regarding the Hardware and Software of Janus, you may like to have a look to references[7], [8] and [9] below. 

The reconfigurable architecture permits Janus to afford different scientific computational applications, as in Physics, in Chemistry or in Biology. So far, the Janus Collaboration has focused its efforts on the study and simulation of spin glasses, paradigm of Complex Systems.

Janus retired in 2020. During the whole Janus’ life, our dedicated machine maintained its edge over the World wide spin glasses simulations, reaching simulation times up to 0.1 seconds of a real experiment, with very large lattice sizes, at very low temperatures and with a huge amount of samples never attempted before. All Janus work has deserved several publications in different outstanding scientific journals, as Physical Review Letters or the Proceeding of the National Academy of Science.  In order to see a complete list of the Janus Collaboration publications, please visit the Publication Section of Janus and JanusII.

References

 
[1] N. Avico, P. Bacilieri, S. Cabasino, N. Cabibbo, L.A. Fernández, G. Fiorentini, A. Lai, M.P. Lombardo, E. Marinari, F. Marzano, P. Paolucci, G. Parisi, J. Pech, F. Rapuano, E. Remiddi, R. Sarno, G. Salina, A. Tarancón, G.M. Todesco, M. Torelli, R. Tripiccione, and W. Tross. From ape to ape-100: From 1 to 100 gflops in lattice gauge theory simulations. Computer Physics Communications, 57(1):285–289, 1989.
 

[2] F. Bodin, Ph. Boucaud, N. Cabibbo, F. Di Carlo, R. De Pietri, F. Di Renzo, W. Errico, H. Kaldass, A. Lonardo, S. de Luca, J. Micheli, V. Morenas, O. Pene, D. Pleiter, N. Paschedag, F. Rapuano, D. Rossetti, L. Sartori, F. Schifano, H. Simma, R. Tripiccione, and P. Vicini. The apenext project. Nuclear Physics B – Proceedings Supplements, 140:176–182, 2005. LATTICE 2004. 

[3] R. Pearson, J. Richardson, D. Toussaint, A Special Purpose Machine for Monte Carlo Simulations, tech. report NSF-ITP-81-139, Inst. Theoretical Physics, Univ. California, Santa Barbara, 1981.

[4] J.H. Condon, A.T. Ogielski, Rev. Sci. Instruments 56 (1985) 1691–1696; A.T. Ogielski, Phys. Rev. B 32 (1985) 7384–7398.

[5] “The RTN Collaboration,” Proc. Computing in High Energy and Nuclear Physics 92 (CHEP), CERN, 1992.

[6] A. Cruz et al., “SUE: A Special Purpose Computer for Spin Glass Models,” Computer Physics Comm., vol. 133, nos. 2–3, 2001, pp. 165–176.

[7] Janus Collaboration, “Ianus: an Adpative FPGA Computer.”, Computing in Science & Engineering, January/February 2006, Volume 8, N 1, p. 41.

[8] Janus Collaboration, “Simulating spin systems on IANUS, an FPGA-based computer.”, Computer Physics Communications 178 (3), p.208-216, (2008).

[9] Janus Collaboration, “JANUS: an FPGA-based System for High Performance Scientific Computing”, Computing in Science & Engineering 11-1, 48-58 (2009).