> On April 8, 2013, 7:41 p.m., Alessandro Presta wrote: > > Weighted PageRank doesn't seem particularly significant in the literature. > > I'd rather keep just one (standard) PageRankBenchmark. > > Also, why would it hurt to keep the current options? We can still use the > > benchmark to test vertex/edge input, edge locality, and different > > VertexEdges implementations, as long as the defaults are the fastest. > > Edge locality especially seems something we should keep in the benchmark, > > since it can show how Giraph behaves with smartly-partitioned data.
The way I see it, we use benchmarks to test the speed of our code when we make some changes or want to try something out. So having the version which also uses weighted edges seems useful to me. Also options from WeightedPageRankBenchmark, like whether to use combiner or not, which partitioner to use and so on, are there to test different parts of our infrastructure. And FastPageRankBenchmark is a simple one, with all the fastest stuff set there already, and which can just tell us how fast a simple application like Page Rank runs on Giraph. I'll put in the locality. - Maja ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/10343/#review18791 ----------------------------------------------------------- On April 8, 2013, 6:23 p.m., Maja Kabiljo wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/10343/ > ----------------------------------------------------------- > > (Updated April 8, 2013, 6:23 p.m.) > > > Review request for giraph. > > > Description > ------- > > Make two benchmarks for page rank: > - WeightedPageRankBenchmark which corresponds to current PageRankBenchmark > (has all the options for testing different stuff), changed the vertex > implementation to use the weights of the edges in the computation > - FastPageRankBenchmark, which we can use to show how fast Giraph really is > :-) It uses integers for vertex ids, floats for page ranks, it's unweighted > and doesn't have all the options from PageRankBenchmark, but just the basic > stuff > > > This addresses bug giraph-543. > https://issues.apache.org/jira/browse/giraph-543 > > > Diffs > ----- > > > giraph-core/src/main/java/org/apache/giraph/benchmark/FastPageRankBenchmark.java > PRE-CREATION > > giraph-core/src/main/java/org/apache/giraph/benchmark/FastPageRankVertex.java > PRE-CREATION > > giraph-core/src/main/java/org/apache/giraph/benchmark/PageRankBenchmark.java > fbb2516 > giraph-core/src/main/java/org/apache/giraph/benchmark/PageRankVertex.java > 3f394b4 > > giraph-core/src/main/java/org/apache/giraph/benchmark/WeightedPageRankBenchmark.java > PRE-CREATION > > giraph-core/src/main/java/org/apache/giraph/benchmark/WeightedPageRankVertex.java > PRE-CREATION > giraph-core/src/main/java/org/apache/giraph/combiner/FloatSumCombiner.java > PRE-CREATION > giraph-core/src/main/java/org/apache/giraph/edge/IntNullArrayEdges.java > PRE-CREATION > > giraph-core/src/main/java/org/apache/giraph/io/formats/PseudoRandomEdgeInputFormat.java > 87cadb4 > > giraph-core/src/main/java/org/apache/giraph/io/formats/PseudoRandomIntNullVertexInputFormat.java > PRE-CREATION > > giraph-core/src/main/java/org/apache/giraph/io/formats/PseudoRandomUtils.java > PRE-CREATION > > giraph-core/src/main/java/org/apache/giraph/io/formats/PseudoRandomVertexInputFormat.java > dca0271 > giraph-core/src/test/java/org/apache/giraph/io/TestJsonBase64Format.java > 0117ce9 > > Diff: https://reviews.apache.org/r/10343/diff/ > > > Testing > ------- > > > Thanks, > > Maja Kabiljo > >
