toc_footers: - Gunrock: GPU Graph Analytics - Gunrock © 2018 The Regents of the University of California.

search: true

full_length: true

Name Of Application

One-paragraph summary of application, written at a high level.

Summary of Results

One or two sentences that summarize "if you had one or two sentences to sum up your whole effort, what would you say". I will copy this directly to the high-level executive summary in the first page of the report. Talk to JDO about this. Write it last, probably.

Summary of Gunrock Implementation

As long as you need. Provide links (say, to papers) where appropriate. What was the approach you took to implementing this on a GPU / in Gunrock? Pseudocode is fine but not necessary. Whatever is clear.

Be specific about how to map the algorithm to Gunrock operators. That is helpful for everyone.

Be specific about what you actually implemented with respect to the entire workflow (most workflows have non-graph components; as a reminder, our goal is implementing single-GPU code on only the graph components where the graph is static).

How To Run This Application on DARPA's DGX-1

Prereqs/input

(e.g., "build Gunrock's dev-refactor branch", "this particular dataset needs to be in this particular directory")

Running the application

include a transcript

Note: This run / these runs need to be on DARPA's DGX-1.

Output

What is output when you run? Output file? JSON? Anything else? How do you extract relevant statistics from the output?

How do you make sure your output is correct/meaningful? (What are you comparing against?)

Performance and Analysis

How do you measure performance? What are the relevant metrics? Runtime? Throughput? Some sort of accuracy/quality metric?

Implementation limitations

e.g.:

Comparison against existing implementations

Comparison is both performance and accuracy/quality.

Performance limitations

e.g., random memory access?

Next Steps

Alternate approaches

If you had an infinite amount of time, is there another way (algorithm/approach) we should consider to implement this?

Gunrock implications

What did we learn about Gunrock? What is hard to use, or slow? What potential Gunrock features would have been helpful in implementing this workflow?

Notes on multi-GPU parallelization

What will be the challenges in parallelizing this to multiple GPUs on the same node?

Can the dataset be effectively divided across multiple GPUs, or must it be replicated?

Notes on dynamic graphs

(Only if appropriate)

Does this workload have a dynamic-graph component? If so, what are the implications of that? How would your implementation change? What support would Gunrock need to add?

Notes on larger datasets

What if the dataset was larger than can fit into GPU memory or the aggregate GPU memory of multiple GPUs on a node? What implications would that have on performance? What support would Gunrock need to add?

Notes on other pieces of this workload

Briefly: What are the important other (non-graph) pieces of this workload? Any thoughts on how we might implement them / what existing approaches/libraries might implement them?