Due by 11:59 pm on Tuesday 11/22

Objectives

1. Practice designing an efficient parallel algorithm
2. Analyze a parallel algorithm in terms of work and span
3. Implement a parallel algorithm using OpenMP

Assignment

As many of you have realized, the problem from Project 1 is better known as tthe RNA Secondary Structure problem, an important problem in biology and genetics. The only difference is that rather than sequences of COMIC-CON fans (G, H, W, T), you work with sequences of RNA bases (A, C, G, U; C pairs with G, A pairs with U). Your task is to design, analyze, and implement a multithreaded algorithm to solve the RNA Secondary Structure problem. You may use a memoized or bottom-up approach, but in either case, your goal is to design an algorithm that makes efficient use of multiple processors.

Once you have designed your multithreaded algorithm, you will need to analyze it. First compute the work T₁(n) and the span T_∞(n) where n is the size of the input set X. You should prove your expressions for T₁ and T_∞ as rigorously as possible. Finally, compute the parallelism and estimate a range of parameters (n and P) for which linear or near-linear speed-up may be expected.

The last part of the project is to implement your RNA Secondary Structure algorithm in C or C++ using OpenMP on UMBC's maya cluster and measure it's performance empirically using 1, 2, 4, and 8 processors (optionally, you may test on 16 processors; most maya nodes have eight cores, but some have 16). You will need to generate test data with variety of different sized input sets and subsets to demonstrate the performance characteristics of your algorithm.

Submission

Your project submission consists of the following:

1. A written report describing the design, analysis, and empirical performance of your algorithm.
2. Multithreaded RNA Secondary Structure code and test data.

The written report must be turned-in on Blackboard as either a DOC, DOCX, or PDF file. Code and test data should be archived in a single zip file and submitted on Blackboard.

Grading

Points will be distributed among design, analysis, and implementation as follows:

• Algorithm Design — 30 points
• Algorithm Analysis — 30 points
• Implementation — 40 points

Hints and Resources

• Be sure to review the project policies.
• Although accounts have already been created for each of you, you will need to complete the account request form so that you can acknowledge the user agreement.
  • Your position is "Undergraduate Student"; your Department, Affiliation, and Address are all "CSEE".
  • Your Role in the Resarch Project is "Member of the research group".
  • The Information on UMBC Faculty/PI Account Sponsor section should be completed with my information: "Chris Marron", "CSEE", "cmarron@umbc.edu", and "x52417".
  • Use "CMSC 441 Project 2" for the Title of the Research Project and "Multithreaded RNA Secondary Structure" for the Abstract.
  • Check the Accept Terms box and click Submit.
  Once you've completed the user agreement, you can login to your account on maya.umbc.edu.
• You should use the Intel compilers (icc and icpc) and the SLURM job management system. To make these available, you will need to load a couple modules in your Linux environment:
  • module load intel/compiler
  • module load slurm
  The first module command loads the Intel compilers into your environment; the second loads SLURM.
• There is an excellent OpenMP Tutorial by Tim Mattson linked from the openmp.org website. There are also video lectures on YouTube.
• An introduction to SLURM on maya is available from the UMBC HPCF website.
• Although I prefer that you use OpenMP, I will accept projects using pthreads so long as you use them in a way which is consistent with the costing model (parralel loops, spawn, sync). If you choose a top-down approach, pthreads may be easier than OpenMP.
• It may turn-out that eight processors are insufficient to fully demonstrate the advanate of the parallel algorithm.