Hi,
We are pleased to announce that UDA3.1 code is released.
We are welcoming more use cases, please contact us for any help you might need, email us to: bigdata@mellanox.com
UDA, Mellanox’s Unstructured Data Accelerator, is an open source acceleration plugin for Hadoop framework.
UDA 3.1 can be found on the following Google code directory: https://code.google.com/p/uda-plugin/
Documentation can be found here: http://www.mellanox.com/page/products_dyn?id=144&product_family=144&mtag=uda
New in UDA3.1:
- Added snappy compression support for intermediate MOF file
- Added resilience fallback to default built-in shuffle in cases of:
- Unsupported user defined compare function
- Unsupported configurations like some compression modes
- Not enough memory to complete job with RDMA shuffle
- General UDA failure
- Enabled UDA log integration into TaskTracker and Reduce task log system
- Support for Apache Hadoop 1.x base MapRecude applications
- Added additional Apache Hadoop and 3rd party Hadoop distributions support
UDA highlights:
- Open-source project
- Integrated support for FDR InfiniBand or 10/40GbE to MapReduce Framework
- Leverage data transfer with RDMA technology and efficient merge-sort algorithm
- Reduce job execution time by up-to 50%
- Better CPU utilization for map and reduce jobs
- Seamless integration to Hadoop framework