1. Potential Research Projects For.  Dissertations.  Theses
Potential Research Projects For  Dissertations  Theses  Undergrad Honor Theses  Short Semester or Summer Projects
Johnnie Baker
Parallel and Associative Computing
Computer Science Department
Kent State University

2. Short Background summary for Projects
NOTE: Projects start on Slide 10

3. Some Advantages of the AP Hardware over MIMD for ATC
Control-type information between processors is eliminated
Communication between the IS and processors occur in constant time
Data communications between processors is deterministic and much faster
Low synchronization overhead
Extremely wide “memory to processor” bandwidth
Rapid I/O
Elimination of shared resource management issues

4. Some Advantages of AP hardware for ATC Hardware (cont)
Elimination of need for additional software to solve difficult problems continuously that is not required for sequential solutions
Dynamic scheduling
Load balancing
Management of shared resources
Preemption management
Synchronization
Cache and memory coherency problems
Management of multi-tasking and multi-threading software
Many others

5. The Sequential RAM model
The tremendous progress in sequential computation over the past 60 years is due to the widely accepted RAM model
A widely accepted single parallel model could result in similar advances in parallel computation.
Sequential programming is a deterministic and predictable process that arises intuitively from the way that programmers solve problems using algorithms.
The sequential programmer typically does not need an indepth understanding of the latest changes in hardware design in order to write efficient programs

6. Ideal Features for a Parallel Model
Parallel programming should be as simple and intuitive as sequential programming
It should describe several features that its target architectures should satisfy, such as easy to use, energy efficient, scalable, and portable.
It is desirable that the model avoid, or at least minimize, software problems like data dependences, race conditions, load balancing, false sharing, deadlock, and non-predictability
The general parallel programmer should not need an in-depth understanding of the latest features in hardware design in order to write efficient programs.

7. The Associative (ASC) model of Parallel Computation
The Associative model satisfies all of the preceding desirable characteristics of an ideal parallel model.
Quite a few software packages have been designed and implemented for this model.
See PDCS article on page 14, with references given.
Several algorithms have been designed for ASC
Graph algorithms: Minimal Spanning Tree & Shortest Path
Several convex hull algorithms & a dynamic convex hull alg.
A string matching algorithm
A wider range of ASC algorithms are needed.

8. Comparison of the AP and MIMD Solutions to ATC
The extra problems that MIMDs have to solve as part of the solution to their original problem requires a lot of work.
Additionally, the extra work that MIMDs have to do to support the integrity and the ACID properties of the dynamic database requires substantially more work
Synchronization, dynamic scheduling, critical sections of code, sorting, indexing, substantially more communications
Due to its dynamic nature, massive updates are being made to this database every 0.5 seconds.
The graphs in Section 8 of the JPDC Reference paper shows the magnitude of this extra work dramatically slows down the time required for the MIMD solution.
The AP graph shows its work increases very slowly since sufficient PEs are assumed so each handles only one plane.

9. Potential Projects

10. Potential Research Projects
Create GPU solutions for our ATC prototype using NVIDIA & new Intel hardware
A combined multicore server plus GPU accelerator solution
Performance of implementations compared with ClearSpeed
Will involve same tasks as the JPDC paper, but Clearspeed timings needs to be rechecked
Have an NVIDIA professional to serve as advisor
Show that MASC, the multiple ASC model, has the desirable properties established for the ASC model in recent JPDC paper using an ATC-type project for timing
Wittaya Chantamas, one of my past Ph.D. students, created a very efficient architecture to support MASC & has built several simulators for this architecture
This project require building an emulator on ClearSpeed to use to test performance
Wittaya could serve as a co-collaborator in project

11. Potential Research Projects (cont. 2)
Implement ATC algorithms from recent JPDC paper on the new ClearSpeed system
Simpler project
Initially would involve learning how to use new system
Ideally, would; also involve obtaining new timings and new graphs, corresponding to ones in the JPDC paper.
Develop a variety of ASC versions of well-known algorithms found in undergraduate and graduate algorithms textbook
Could involve just one algorithm or several
Needed to demonstrate usefulness of ASC model and if using ASC to teach parallelism to undergraduates
Algorithms can be published, esp. if implemented and performance compared to a MIMD implementation
Often ASC algorithm can be obtained by modifying an existing sequential algorithm

12. Potential Research Projects (cont. 3)
Many choices of types of algorithms possible, including
Sorts and selections: biotonic sort, merge sort, quick sort
Graph algorithms (strongly connected components, alternate spanning tree or shortest path algorithms like Kruskal’s algorithms
Classic algorithms like FFT
Alternate string matching like Rabin-Karp, Knuth-Morris-Pratt, etc,
Dynamic Programming like longest common subsequence
Greedy algorithms like Huffman code
Computational geometry algorithms
Investigating a theoretical architecture to execute PRAM algorithms with the same running time complexity
Background covered in PDA course or Akl’s textbook
Current cost to support PRAM algorithms is O(lg n)

13. Potential Research Projects (cont. 4)
Implement a variant of our current ATC ClearSpeed version that uses transponder information from the aircraft as much as possible (as is done by FAA).
One student currently considering this problem
Applying our implementation of ATC on ClearSpeed to UASs (Unmanned Aircraft Systems) or drones
Use our AP tracking techniques for aircraft to replace the inefficient Huffman filter normally used in ATC to “smooth the tracking of aircraft”.
Wittaya Chantamas is interested in working on this with me, but we may need a undergraduate or graduate student to join project to implement and test
Based on a MA thesis of K.M. Liu, “Composition of Kalman and Heuristic Tracking Algorithms for Air Traffic Control.

14. Potential Research Projects (cont. 5)
The source of several projects would be to apply our associative ATC techniques to UASs (Unmanned Aircraft Systems) or drones.
FAA is being forced to open up air space to UASs in 2015.
Will have major use in agriculture, defense systems, patrolling national borders, monitoring wildlife both to protect animals (elephants) or do more complete counting and monitoring of them
Can be used to monitor large national parks, etc to detect fires before they get out of hand.
All UASs can easily take video images of area they monitor.