Bioinformatics on network processors

So far network processors units (NPUs) have been exclusively used in networking systems. In this field they form an attractive solution as they offer high throughput, full programmability and are expected to scale with link speed. We observe that bio-computing fields like DNA processing share many of the properties and problems of high-speed networks. At the same time, we observe that DNA processing suffers from sub-optimal performance when implemented on general-purpose processors. Many of the problems are caused by the fact that the potential parallelism in DNA processing cannot be exploited by general purpose processors. For this reason, we evaluate the suitability of NPUs for implementing well-known DNA processing algorithms, such as `Blast'. To achieve a realistic performance measure, the first (parallelisable) stage of Blast has been implemented on an Intel IXP1200 network processor and used to process realistic queries on the DNA of a Zebrafish. It is shown that the performance of a 232MHz IXP1200 is comparable to that of a 1.8GHz Pentium.


On the feasibility of using NPUs for DNA processing .
(NP3, Madrid, February 2004.) [PDF version]

(A slightly modified version of this paper will be published as Chapter 10 in "Network Processor Design, Vol. 3", Morgan Kaufmann, pp. 10-1 -- 10-14, 2004.)