Any Hope for the Sun Worshippers?
At Ctest Labs, we have (or had?) really big plans for the Sun Ultrasparc T2 and maybe Niagra 3. These are very nice boxes and fit our multicore needs just fine. But now that Oracle is acquiring Sun, what will become of the Ultrasparc T1, T2, and that whole line of multicore computers that had CMT?We already have a T1 as part of our Pantheon cluster at Ctest. It provides us with 8 Sparc V9 cores. The V9 cores are 64 bit technology. Each core has a L1 cache. This diagram shows simple layout of the Ultrasparc chip:
The eight cores all share a single floating-point unit. Each core has a six-stage pipleline. Our T1 keeps company with two dual processor Sun Blade 2000s, a Quad-Processor Sun Ultra 80, a few Sun Ultra 60s, a dual-core Opteron, and two Cell processors by way of the PS3. Together these boxes make up our Pantheon cluster. We tie all this together with MPI and PVM. Although PVM is not as hot as it once was, it works, its simple, and we like the dynamic node management. We do allot of our big bang theory stuff with MPI.
We were hoping to add the Sun Ultrasparc T2 to our Pantheon mix this year. The configuration we had in mind has 8 cores running at 1.4 Ghz, 64 hardware threads, and 4 MB of level 2 cache. But its not entirely clear what Oracle's plans are for the hardware division at Sun. Conventional wisdom says that Oracle will continue to support and feed the growth of Java. That's a good thing. Oracle also has a vested interest in keeping Solaris alive and well. That's another good thing. At Ctest, we use Solaris both on X86 and Ultrasparc technology. Solaris has quite nice support for multithreading and parallel programming. So with this Oracle thing now underway, we think the software from Sun is safe, but what about the hardware? It would be a tragedy for the advancement of multicore technology if the Sun hardware plans go the way of the dinosaur. That would also dampen our spirits here at Ctest Labs where Sun technology is a major part of our multithreading and parallelization strategies at the hardware level. Sun's T1000 and T2000 servers definitely brought a different flavor of multicore to the table, and as far as we're concerned Sun's C++ compiler's multithread support, OpenMP support are absolutely top rate. Sure, we can get by on Opterons from AMD, but there's a special place in our hearts for Sun. We grew up with and cut our teeth on Sun technology. We can only hope that Oracle does the right thing with the hardware division at Sun. But no matter how things turn out with Sun and Oracle, we will keep our Sun equipment running until the wheels fall off, third-party suppliers run out of parts, and eBay goes silent.