Apple’s iOS, Google’s Android, and MeeGo from Intel and Nokia will continue to butt heads in various platforms from smart phones to tablets, providing a sizable number of developers with a range of target platforms. Of more interest to embedded developers is the use of these platforms to augment other application hardware and software using devices like the iPad and Droid as control panels.
Developers will be hard pressed to match the price and options of the plethora of iPad alternatives or the iPad itself. Look for more products to incorporate smart-phone-enabled Web servers and support applications.
Last year, myriad multicore designs were released with advanced software emulation support that allowed developers to get started before the chips were delivered. This will be even more popular this year as developers especially like the exposure of system internals unavailable in the real hardware.
Multicore Programming
The approaches to tackling multicore platforms are as varied as the platforms themselves. Single-chip, symmetric multiprocessing (SMP) multicore is readily addressed using existing tools that are typically amenable to partitioning, especially when coupled with virtualization. In these cases, the number of cores is an advantage.
Clusters of multicore processors kick the number of cores into the thousands. Here, OpenMP, MPI, and OpenCL come into play (see “Parallel Programming Is Here To Stay”). These tools have been moving out of academia and into the commercial space, and this year will see that trend accelerate. Intel’s Intel Parallel Studio XE and Cluster Studio are just a couple of the new tools developers will be able to utilize (see “Parallel Programming Platform Targets Linux And Windows”).
Not surprisingly, students regularly compete with supercomputers that are now more affordable (see “What Kind Of Supercomputer Can You Build With 26 Amps?”). This is especially true when graphics processing units (GPUs) are added to the mix, as they have been for a number of years.
The big difference lately is the number of commercial applications taking advantage of the hundreds of cores in a GPU. Many systems match multiple GPUs to a CPU. Rack-mount GPU systems are available as standard fare from vendors like NVidia, Supermicro, Appro, and AMBX Servers.
Developers will need to use parallel programming tools for platforms like SeaMicro’s SM10000 (see “10U Rack Packs 512 Atoms”). The SM10000 holds 512 1.6-GHz Z530 Atom processors and 1 Tbyte of DRAM.
Tilera’s Tile-GX100 (Fig. 2) puts 100 very long-instruction-word (VLIW), 64-bit cores on a single chip. The mPIPE wire speed packet processing system acts as a front end for up to 32 Gigabit Ethernet ports. Developers can partition SMP regions and even run individual cores with zero overhead Linux support.
Some developers are using graphical programming environments like National Instruments’ LabVIEW as well as matrix-oriented tools like the MathWorks’ MATLAB as alternatives for parallel processing. Both support GPUs in addition to multicore platforms.
Securing Applications
Counterfeiting is likely to drive developers more than security given the limited appreciation for secure practices in most embedded development environments. Firewalls, encrypted communication links, and authentication mechanisms are readily available these days, but coordinated use is often more complex than the underlying application. The challenge remains in exposing the payoff in the long run.
Continue on next page