The Nvidia CUDA development kit has been updated to version 4.1.
DAS-4/ASTRON has succesfully correlated three hours of data coming from 288 LOFAR antennas, possibly setting a world record.
The GPU page documents how the OpenCL implementations on DAS-4 can be used.
DAS-4/VU achieved the 16th place on the Graph500 list. A very good result, as all higher-ranked systems are significantly bigger.
DAS-4 supports the OpenNebula package for Cloud computing.
Each DAS-4 cluster has both an internal 1 Gbit/s Ethernet and a QDR InfiniBand network. The Ethernet network (with IP addresses 10.141.<site-id>.<node-id>) is mainly used for management of the cluster, but can also be used by applications.
The best performance is obtained using the InfiniBand network, which offers low latency and throughputs exceeding 20 Gbit/s (depending on the networking API used). The InfiniBand network can be accessed both via a fast native interface, typically through MPI, but also by means of the regular IP layer, using IP addresses 10.149.<site-id>.<node-id>.
For external connectivity, each cluster has a 1 or 10 Gbit/s ethernet connection to the local university backbone; the head node acts as a router for this purpose.
In addition, DAS-4 compute nodes can communicate over a
very efficient Wide-Area network called StarPlane, which
is based on dedicated 10 Gbit/s light paths provided by SURFnet.
To use this network, simply specify InfiniBand-based IP addresses (10.149).
Routes to external InfiniBand IP addresses are set up to go
through a local WAN router, which routes between InfiniBand
and the SURFnet 10 Gbit/s lightpaths.
For details on the wide-area topology, see file
/cm/shared/package/StarPlane/overview on DAS-4.