Modern Computational Techniques for the HMMER Sequence Analysis

Table 1

Performance comparison among different software approaches.

Acceleration strategies	Supporting software/package	Accelerated programs	Datasets	Hardware environment	Achieved speedup
Instruction-level parallelism	SSE2 Instructions	hmmpfam, hmmsearch	Pfam and nr databases	2.66 GHz Intel Xeon processor with 2.5 GB of memory	1.2x~1.3x [9]
Shared memory parallelism	OpenMP	hmmpfam	600 HMM profiles and 250 sequences	16 x86 3.0 GHz processors, 32 MB L4 cache shared among 4 CPUs, 4 MB L3 cache, 8 GB of memory	14x [10]
	EARTH	hmmpfam	50 HMM profiles and 38192 sequences	A cluster that consists of 128 nodes, each with two 500 MHz Pentium III processors	222.8x [11]
Distributed memory parallelism	PVM	hmmsearch	1 HMM profile and 100 MB of nr database	A cluster with 4 nodes, each node consists of two 2.66 GHz Intel Xeon processors with 2.5 GB memory per node	4.56x [9]
	MPI	hmmsearch			5.90x [9]
	MPI + I/O optimizations	hmmsearch	One 236-state HMM profile and nr database	A cluster that consists of 1056 nodes, each equipped with two 3.2 GHz Intel Xeon processors, 2 GB RAM	221x [12]
		hmmpfam	1.6 GB of Pfam database and nr databases		328x [12]
Heterogeneous approach	MPI  +  SSE2	hmmsearch	1 HMM profile and 100 MB of nr database	A cluster with 4 nodes, each node consists of two 2.66 GHz Intel Xeon processors with 2.5 GB memory per node	7.71x [13]