UpPipe: A Novel Pipeline Management on In-Memory Processors for RNA-seq Quantification
DescriptionA key overhead of RNA-seq quantification is mapping RNA reads to transcriptome. Besides, performance of RNA-seq quantification is limited by von-Neumann bottleneck. As the first publicly commercial PIM system, UPMEM DPU, is proposed, PIM becomes a promising solution to resolve von-Neumann bottleneck. We aim to improve the performance of RNA-seq quantification by exploiting strengths of UPMEM. We propose a novel DPU-based pipeline design ``UpPipe'' with considering DPU's unique constraints. In UpPipe, a transcriptome allocation policy is proposed considering the limited MRAM capacity of DPU. Furthermore, pipeline management is presented to efficiently map RNA reads to the distributed transcriptome over DPUs.
TimeTuesday, July 11th4:55pm - 5:10pm PDT
Location3010, 3rd Floor
In-memory and Near-memory Computing Circuits