We present results of scaling an ab initio motif family identification system, Dragon Motif Finder (DMF), to 65,536 processor cores of IBM Blue Gene/P. DMF seeks groups of mutually similar polynucleotide patterns within a set of genomic sequences and builds various motif families from them. Such information is of relevance to many problems in life sciences. Prior attempts to scale such ab initio motif-finding algorithms achieved limited success. We solve the scalability issues using a combination of mixed-mode MPI-OpenMP parallel programming, master-slave work assignment, multi-level workload distribution, multi-level MPI collectives, and serial optimizations. While the scalability of our algorithm was excellent (94% parallel efficiency on 65,536 cores relative to 256 cores on a modest-size problem), the final speedup with respect to the original serial code exceeded 250,000 when serial optimizations are included. This enabled us to carry out many large-scale ab initio motif-finding simulations in a few hours while the original serial code would have needed decades of execution time.

Effective scaling and maximal throughput for clusters is an ongoing issue for computational workloads. We will discuss an approach to workload distribution which is data-centric, rather than process-centric. Data is moved to nodes ahead of computation. The workload management is handled as usual, except that the presence of data at nodes is used to trigger process scheduling. We will discuss how to create workflows which are data-activated and how an asynchronous pipeline can be established, allowing file serving latency to be hidden.In addition our RepliCator solution provides broadcasting to all nodes in a cluster simultaneously, substantially reducing data transfer times. This can be highly effective for throughput problems. We will present benchmark results obtained on clusters with several hundred cpus. Speedups in overall processing of factors of 2.5 to 4.5 have been observed using the combination of data-activated processing and data broadcasting provided by RepliCator data transfer management software.