All, Please check this issue:
https://issues.apache.org/jira/browse/HAMA-486 On Fri, Nov 25, 2011 at 11:16 AM, Chia-Hung Lin <[email protected]> wrote: > +1 for pluggable feature/ hadoop's counter arch > > On 25 November 2011 01:05, Thomas Jungblut > <[email protected]> wrote: >> Great, I open an issue for this counter stuff. >> >> 2011/11/24 Edward J. Yoon <[email protected]> >> >>> >> I think it would be good as a 'pluggable' feature as this counting >>> could>> impact performances. >>> and >>> >>> > It is difficult to say whether we actually need this.* >>> > *In case of SSSP it is useful, but as Tommaso says it can degrade >>> >>> I agree with you. If possible, let's provide as a optional feature. >>> >>> > What do you think of a similar architecture of Counters like in Hadoop? >>> >>> Good. >>> >>> Currently, our Job client reports only shows the total number of >>> supersteps. I think that the Job reports needs to be strengthened. :-) >>> >>> On Thu, Nov 24, 2011 at 11:05 PM, Thomas Jungblut >>> <[email protected]> wrote: >>> > It is difficult to say whether we actually need this.* >>> > *In case of SSSP it is useful, but as Tommaso says it can degrade >>> > performance for every other BSP by increasing the cost of the sync. >>> > >>> > What do you think of a similar architecture of Counters like in Hadoop? >>> > This could contain even more information, e.G. the number of bytes >>> > transferred etc. >>> > >>> > 2011/11/24 Tommaso Teofili <[email protected]> >>> > >>> >> Hello Edward, >>> >> >>> >> I think it would be good as a 'pluggable' feature as this counting could >>> >> impact performances. >>> >> So it'd be useful when explicitly "debugging" while it should be, in my >>> >> opinion, disabled by default. >>> >> >>> >> My 2 cents. >>> >> Tommaso >>> >> >>> >> >>> >> 2011/11/24 Edward J. Yoon <[email protected]> >>> >> >>> >> > Hi all, >>> >> > >>> >> > I'd like to add counting feature to get the total number of message >>> >> > exchanges during one synchronization among BSP tasks. Before leave >>> >> > barrier in sync() method, each task can report their counts. >>> >> > >>> >> > This information will be useful to user and, some applications (for >>> >> > example, the sync cost of current SSSP example can be reduced to 1/3). >>> >> > Should we add this feature? >>> >> > >>> >> > -- >>> >> > Best Regards, Edward J. Yoon >>> >> > @eddieyoon >>> >> > >>> >> >>> > >>> > >>> > >>> > -- >>> > Thomas Jungblut >>> > Berlin <[email protected]> >>> > >>> >>> >>> >>> -- >>> Best Regards, Edward J. Yoon >>> @eddieyoon >>> >> >> >> >> -- >> Thomas Jungblut >> Berlin <[email protected]> >> -- Best Regards, Edward J. Yoon @eddieyoon
