Hone: Performance Benchmarks for Scaling Down Hadoop on Shared-Memory Systems

K. Ashwin Kumar from the University of Maryland, College Park presents a prototype runtime called Hone for executing Hadoop applications on multi-core, shared-memory machines. The dataset likely contains performance metrics comparing Hone's execution on a single server against Hadoop running in pseudo-distributed mode and on a 16-node cluster. The paper demonstrates that for datasets fitting into a single machine's memory, Hone can be substantially faster.

Use Cases

• Benchmarking MapReduce algorithm performance based on runtime environment comparisons mentioned in the description
• Evaluating the efficiency of shared-memory systems for analytics tasks based on the prototype's experimental results
• Comparing single-server versus multi-node cluster execution for varying dataset sizes as discussed in the paper

Strengths

• Benchmarks include a direct comparison to a 16-node Hadoop cluster, providing a concrete performance scale
• Focus on a specific, re-evaluated trend in analytics involving smaller, more refined datasets
• Prototype is API compatible with Hadoop, allowing for testing with existing applications

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Last update date is unknown; freshness unverified

Provenance

Source

University of Maryland, College Park

Collection Method

Likely experimental results from a research prototype.

Time Range

null

Freshness

Last updated date is unknown.

Geography

null