Note: These tutorials are incomplete. More complete versions are being made available for our members. Sign up for free.

Heteregenous and Reconfigurable Computing for Bioinformatics

The hardware solutions considered so far attempt to process large NGS libraries by using the combined power of many processors, but the underlying processor architecture was Intel-based CPUs. Heterogeneous computing is a new paradigm, where

The computing paradigm presented so far considered only one type of processor.

Has similar challenges as Condor, but at the hardware level.

The hardware solutions considered so far attempt to process large NGS libraries by using the combined power of many processors, but the underlying processor architecture remains the same (typically CPUs designed by Intel). In contrast, in hardware accelerator-based solutions, the basic processor units are replaced by specialized computer chips custom-designed to address specific issues. FPGAs and hardware-accelerators are non-intuitive solutions for bioinformatians experienced with developing software algorithms. Therefore, we will first explain, where they fit in the big picture, with a simple example. A computer can solve a problem in two ways - (i) software solution – the microprocessor chip gives the programmer a small number of hardware-implemented ‘operations’, and the programmer builds more complex programs using those operations as basic units, (ii) hardware solution – the chip itself is sophisticated enough to solve a complex problem using the circuits prebuilt in it.

Imagine a computer has a processor chip that can only add numbers. A prograamer can still use the chip for multiplication. He needs to write a software program that uses the adder many times. He may even write a more sophisticated multiplication algorithm that looks up a table to speed up the multiplication. That is how a software solution for multiplication works.

Instead, a hardware based solution has the multiplication algorithm prebuilt into it using transistors. It adds an extra ‘operation’ for multiplication that the software programmer invokes, when he wants to multiply two numbers. This frees up software programmer’s time to solve more complex problems.

Relative merits and demerits of two approaches can be understood from the above example. Hardware-based solutions can run orders of magnitude faster than software programs. In fact, more complex the task, higher is the improvement in speed. On the other hand, software-based approaches are easy to implement. In addition, a software-based solution is easy to reconfigure, but a silicon chip custom-designed for one algorithm cannot be modified, if a better algorithm is discovered later.

To give you an idea about how much increase in speed picocomputing The described hardware system demonstrates a 250X speedup versus the original BFAST software version, and a 31X speedup versus Bowtie. Also, the hardware system is more sensitive than Bowtie, which aligns approximately 80% of the short reads, as compared to the 91% aligned by the hardware. Embedded applications run very fast, but we do not know of any application to compare.

Power savings

A second advantage is savings in power. Even with equal costs, another big advantage of FPGA-based solution is its power usage.

GPU

Other Microprocessors such as ARM, MIPS

custom ASIC If bioinformatics has a large market, eventually it may be possible to design custom semiconductor chips to solve various bioinformatics tasks efficiently. A significant fraction of

Design and fabrication of ASICs is usually very expensive and most of that expense comes as fixed cost. Therefore, ASICs are not built until the investors are convinced that they will find a big and steady market for the product.

FPGA

Reconfigurability of FPGA chips comes at a cost. They are generally slower than another class of digital processors known as ASICs, or application specific integrated circuits. As the name suggests, ASICs are custom designed and built for specific tasks and cannot be reconfigured. Intel’s Pentium microprocessor chip is possibly the most famous ASIC, but not the only one. Millions of ASICs get used in every gadget and machine ranging from cell phone, microwave, camera, TV, car and many other nook and corners of our society.