Tuesday 6th of february 2007 - 4.30 PM
SISSA Room E

Antonio Lagana
Dept. of Chemistry
University of Perugia

From the italian grid to a virtual organization of the european grid for molecular sciences

The design and the implementation of grid computing infrastructures and middleware are making increasingly possible to map complex computational chemical applications into an appropriate composition of procedures suitable for distribution on the grid. Such a distribution is aimed at exploiting both the high throughput of grid platforms and the possibility of establishing virtual organizations (VO) allowing an articulation of the work in a cooperative fashion.

The joint effort of some chemistry and computer science laboratories in Perugia has fostered (Within the National Grid project GRID.it, first, and within a COST Chemistry european Action, afterwards) the design of a cooperative workflow for simulating molecular science problems using rigorous theoretical approaches. As a result, a grid enabled molecular simulator (GEMS) has been assembled by gathering the competences of several European research laboratories and its prototype version (GEMS.0) has been implemented as a demo on the European Grid infrastructure of EGEE. On this basis the COMPCHEM VO has been established in EGEE in order to foster the development of accurate realistic Molecular Science. At present work is being carried out to further extend the VO and develop related tools.

GEMS.0 has been already used to simulate molecular beam experiments and to carry out extended computational campaigns of the properties of some gas phase reactions relevant to the modeling of some modern technologies. The structuring of the computations for the grid and the results obtained using different theoretical approaches will be discussed for a case study.

Friday 17th of November - 14:00
SISSA Room A

Giovanni Pau

TCP Libra: Maintaining Fairness at High Speed


The majority of Internet users rely on the Transmission Control Protocol (TCP) to download large data and multimedia files from remote servers (i.e. P2P file sharing). The current TCP new reno is aging and is becoming obsolete for today's high speed networks. In particular, TCP suffers from scalability and RTT fairness issues. For instance, when session round-trip-times (RTTs) radically differ, the share (of the bottleneck link) may be anything but fair; and when used in long distance high gigabit networks the link utilization is far from optimal. This motivates us to explore a new TCP, TCP Libra, which guarantees fair sharing regardless of RTT while achieving almost 100 % efficiency in high-speed networks. The key element of TCP Libra is the unique window adjustment algorithm that provably leads to an RTT-independent throughput for each flow, thus converging to fair share. We position TCP Libra in a non-linear optimization framework, proving that it provides fairness (in the sense of max-min fairness of potential delays) among TCP flows that share the same bottleneck link. Equally important is friendliness of Libra towards legacy TCP. TCP Libra features RTT fairness, TCP New Reno friendliness and the ability to scale in multi-gigabit networks. TCP Libra is has been implemented for Linux 2.6.15. An NS 2 simulation model is also available. Via analytic modeling, simulation, and measurement experiments, we show that TCP Libra achieves top efficiency while maintaining fairness and friendliness to TCP New Reno. A comparison with other TCP versions that have been reported as RTT-fair in literature is also carried out.

Tuesday, 21 June 2005 - 14:30
SISSA Room A

Nikola Pavkovic
( Rudjer Boskovic Institute Bijenicka 54 HR-10000 Zagreb )

"DCC tool for cluster computing"


Debian Cluster Components is a complete software package that eases installation and administration of Debian Linux based computation clusters. It consists of various software components (LDAP, TORQUE, System Installation Suite, Ganglia, MPI & PVM libraries...), integrated in a way that makes installation, administration and using the computation cluster very comfortable. The presentation will cover the DCC project overview and a short demonstration of DCC/Live, the first educational tool for parallel programming in form of a virtual Linux cluster on a LiveCD. For more information check the DCC web site http://dcc.irb.hr/
Presentation of the project

Monday, 22 November 2004, 10.00 SISSA, National Simulation Centre INFM/Democritos.

"HPSC: a Technical Overview of the Joint Proposal Sun Microsystems - AMD"

Wednesday, 3 March 2004, 14.30 SISSA Room A

David Welton

"Tcl and C"
Combining Tcl Scripting with C
The Best of Both Worlds

Thursday, 4 March 2004, 14.30 SISSA Room A

Axel Kohlmeyer

"64-Bit versus 32-Bit CPUs in Scientific Computing"

Friday, 16 January 2004, 14.30 SISSA Room A

Maurizio Morosini/ Sergio Re (SGI)

"Sistemi SGI per il calcolo ad alte prestazioni e per il data Storage"

Tuesday, 14 October 2003, 10.00 SISSA Room A

Stefano Cozzini, Moreno Baricevic, Marco Talpo

"Risorse computazionali della Sissa"


Discuteremo delle diverse risorse computazionali a disposizione degli utenti in questo momento e di quello che arrivera' a breve.
Presenteremo anche le due nuove macchine openMosix (Somaro e Mulo) comprate su fondi di ricerca dei gruppi Maritan e Sorella con il contributo sostanziale del CED.

Il programma di massima e' il seguente:

ore 10.00: Risorse computazionali alla Sissa: stato dell'arte e futuro (Stefano Cozzini)

ore 10.30: Benchmarks su codici di produzione (Davide Ceresoli)

ore 10.40: Introduzione a openMosix e utilizzo del cluster openMosix Mulo/Somaro (Moreno Baricevic)

ore 11.10: discussione su scenari futuri del calcolo ad alte prestazioni in Sissa.

Friday, 23 May 2003, 11.00 SISSA Room A

Fabrizio Coccetti (SLAC Stanford)

"TCP Stacks on High Speed Links"


Rapid and reliable data transport becomes a challenge on multi-Gbps links. The Stanford Linear Accelerator Center has been part of an effort to break the Internet 2 Land Speed Record and demonstrate how Reno TCP can reach high throughput if well tuned. This presentation shows the performance of Reno TCP and of new TCP stacks on unloaded high speed links and in production networks.




misure per l'I2 land speed record (link1, link2)

Test di nuovi stack TCP su link in produzione (questi risultati sono nuovissimi e ci sono test ancora in corso)

Presentazione per I2 preparata da Les Cottrell e Fabrizio Coccetti per il meeting dello scorso aprile a Washington.

last modified: 02.02.2007