usecase_hedges: A C++ repository from UPMEM

usecase_hedges

DNA error-correcting code pipeline (inner -> HEDGES, outer -> RS (Reed-Solomon)) running on UPMEM PIM DPU.

This project is based on the original HEDGES Research Article. (Williamm H.Press, John A.Hawkins and all, Texas University, June 6, 2020, https://www.pnas.org/doi/full/10.1073/pnas.2004821117).

The source code is based on the original HEDGES implementation (https://github.com/whpress/hedges).

     ____________________________________________________________
    |   ___________      _______________      _______________    |
    |  |           |    | outer encoder |    | inner encoder |   |
    |  | synthetic |--->|     (R.S)     |--->|   (HEDGES)    |   |
    |  | data      |    |     [CPU]     |    |   [CPU/DPU]   |   |
    |  |___________|    |_______________|    |_______________|   |
    |                                               |            |
    |                                          [A.C.G.T...]      |
    |                                      (encoded DNA strands) |
    |   ___________      _______________      ______|________    |
    |  | synthetic |    | outer decoder |    | inner decoder |   |
    |  | data      |<---|    (R.S)      |<---|   (HEDGES)    |   |
    |  | decoded   |    |    [CPU]      |    |   [CPU/DPU]   |   |
    |  |___________|    |_______________|    |_______________|   |
    |____________________________________________________________|

The current implementation supports only one DPU as it consists of a POC.

Only the HEDGES inner decoder has been implemented on DPU yet.

Theses performances are described in the next section.

HEDGES inner decoder performances summary

Performances with 1 DPU are measured on real HW. Performances with 2560 DPUs (one server) are estimated based on the one-DPU performance, assuming linear scaling.

CR	srate	drate	irate	strands/DPU run	time (sec/DPU)	decoding throughput with one DPU (seq/sec)	(estimated) decoding throughput with 2560 DPUs (seq/sec)	DPU pipeline efficiency (%)
0.5	0.03105	0.00945	0.00585	510	29.72	17	43,925	69
0.33	0.03105	0.00945	0.00585	510	8.31	61.4	157,092	62
0.25	0.03105	0.00945	0.00585	510	4.37	116.6	298,569	69

usage

install dependencies

The HEDGES project is built on top of nr3python, numpy1.13 and python2.7.

./install.sh

build project

make clean && make

run on DPU HW

In hedges_pipeline.py, make sure that the DPU mode is enabled for the decoding step.

test_dpu_decoder = True