ALPS: Software Framework for Scheduling Parallel Computations with Application to Parallel Space-Time Adaptive Processing

Kyusoon Lee and Adam W. Bojanczyk Cornell University Ithaca, NY, 14850 {KL224,AWB8}@cornell.edu

High Performance Embedded Computing (HPEC) Workshop 18 20 September 2007

ALPS Software Framework

/* This is a .cpp file of the partially adaptive STAP methods that the user had in mind. This is optimized for minimum latency or maximum throughput on a given parallel machine. */ #include <alps.h> ...

Users idea on Parallel Computations (e.g. STAP)

ALPS CodeGen

ALPS Library

Various STAP heuristics constructed from ALPS building blocks

ALPS Scheduler
Find optimal configuration that minimizes overall makespan or maximizes throughput

Task Graph

ALPS Benchmarker & Modeler

Modeling execution times of basic building blocks under various configurations

Building execution time models using incremental weighted least-squares

Difficulties Measurements are not quite reliable Number of variables in the model is large
0.7 0.6 0.5

Relative errors from weighted least-square

0.4 0.3 0.2 0.1 00 0.1 0.2 0.3 0.4 0.5

0.6

Relative errors from ordinary least-square

be tte 30 be r % t t 40 e be r % tte be r tte r

26.6% error from ordinary least-square

20 %

10 %

7.5% error from INV-weighted least-square

0.7

Scheduling tree-shaped task graphs on parallel computers exploiting mixed-parallelism

Difficulties Communication cost not negligible Arbitrary computation costs Trade-off between task-parallelism and data-parallelism
Throughput (1/sec)

Makespan (sec)

30

estimated achieved

0.1

estimated achieved

20 0.05 10

Dataparallelism only

Taskparallelism preferred

Our scheduling algorithm

Dataparallelism only

Taskparallelism preferred

Our scheduling algorithm