摘要： 对BLAST 启发式算法的实现做了优化：引入批处理的概念、并对整个库文件建立哈希表，实现了I/O 延迟掩藏，提高了整个比对过程的速度，同时降低了内存消耗。优化的算法有利于并行化的实现：在并行系统中，将库文件广播到各个计算节点，由节点在局部分别建立哈希表。然后将查询文件分割发送到各个计算节点并行比对，计算结果可以在节点直接输出，不需要主结点收集，减少了通信开销。
关键词:
BLAST；批处理；并行；集群；曙光4000A
Abstract: Through batch processing and building a database based hash table, optimized heuristic algorithm BLAST overlaps computation with I/O and speedup the process of alignment, besides it reduces the requirement of memory. Optimized BLAST is suitable to be parallelized. The whole database is broadcast to each computing node and database based hash tables are built. Then, query sequence segments are sent to each computing node and align with database in computing node. The cost of communication is reduced because the local alignment results are not necessary to be collected to sort.
Key words:
BLAST; Batch processing; Parallel; Cluster; Dawning 4000A
谭光明，徐琳，周幼英，冯圣中，孙凝晖. 基于曙光 4000A 的BLAST 并行算法[J]. 计算机工程, 2006, 32(10): 45-46,49.
TAN Guangming, XU Lin, ZHOU Youying, FENG Shengzhong, SUN Ninghu. Exploiting Parallelization of BLAST on Dawning 4000A[J]. Computer Engineering, 2006, 32(10): 45-46,49.