Roadmap

• Make it easier to send counter info from IterativeReduce apps
• Add YARN web interface history server
• Multiple file input support
• Make log4j.properties more flexible
• Refactor API to be more intuitive, follow Hadoop patters
• Refactor Master/Workers for single compute() calls
• Refactor Updateable into Writables
• Fix Input / RecordParser to work more closely with native Hadoop InputFormats
• Fix RPC to allow for:
  • More flexibility (random updates, non-blocking updates)
  • More intelligent about workers (node checkin, dropout, etc.)
• Provide bindings for Java-GPU
• Provide JNI shim so workers can be in C++