>> the big thing for BSP engines is checkpointing, where the sync point becomes >> the thing you can checkpoint.
Heavily agree with you! On Wed, Jun 8, 2011 at 8:15 PM, Steve Loughran <[email protected]> wrote: > On 06/08/2011 06:22 AM, Edward J. Yoon wrote: >> >> Hi all, >> >> I would suggest that we have to focus on performance improvements and >> basic communication parts of BSP framework. So I've little edited our >> roadmap page. >> >> http://wiki.apache.org/hama/RoadMap >> >> If you have any comments or other opinions, let's discuss here. >> >> Thanks. > > I'd recommend targeting running the fault tolerant code under the MR-279 > engine (hadoop 0.22 or .23?), as this will deploy more than just the MR > engine, and handle liveness -so you don't need to replicate core modules. > > the big thing for BSP engines is checkpointing, where the sync point becomes > the thing you can checkpoint. > -- Best Regards, Edward J. Yoon @eddieyoon
