Trestles, a supercomputer launched earlier this year by the San Diego Supercomputer Center (SDSC) at the University of California, San Diego, is proving itself as a valuable resource for researchers across a wide range of disciplines, from astrophysics to molecular dynamics, who need access to computational resources with rapid turnaround.
In the seven months since its launch, more than 50 separate research projects have been granted time on Trestles. As the system is targeted to serve a large number of users, project allocations are capped at 1.5M SUs (service units, or single processor hours) annually on the new SDSC system.
With 10,368 processor cores, a peak speed of 100 teraflop/s, 20 terabytes memory, and 39 terabytes of flash memory, Trestles is one of several new HPC (high-performance computing) systems at SDSC. The center is pioneering the use of flash-based memory, common in much smaller devices such as mobile phones and laptop computers but relatively new for supercomputers, which typically rely on slower spinning disk technology. "Flash disks can read data as much as 100 times faster than spinning disk, write data faster, and are more energy-efficient and reliable," said Allan Snavely, associate director of SDSC and co-PI for the new system. "Trestles uses 120GB flash drives in each node, and users have already demonstrated substantial performance improvements for many applications compared to spinning disk."
Trestles is appropriately named because it is serving as a bridge between SDSC's current resources such as Dash, and Gordon, a much larger data-intensive system scheduled for deployment in late 2011/early 2012. Like Dash and the upcoming Gordon system, Trestles is available to users of the NSF's TeraGrid, the nation's largest open-access scientific discovery infrastructure. TeraGrid is transitioning this summer to a new phase called XSEDE (Extreme Science and Engineering Discovery Environment), the result of a new multi-year NSF award.
Trestles is among the five largest supercomputers in the TeraGrid/XSEDE repertoire, and is backed by SDSC's Advanced User Support group, which has established key benchmarks to accelerate user applications and assist users in optimizing applications. Trestles provides the Portland Group (PGI), Intel, and GNU compilers. MPI implementations provided are mvapich2 and openMPI. The majority of libraries are compiled for PGI and mvapich2 support, which have been shown to give optimal performance on the system. As such, they are loaded by default in the login shell. Intel and GNU compilers should only be used for compatibility, if necessary. Similarly, mvapich2 should be used over OpenMPI whenever possible. For more information, click here.