Canceling this RC vote. --- me: hi Suraj: I was reviewing code for async send of messages I think we are double counting the messages sent probably a wrong time to say is it a minor issue for this release? me: double counting? Suraj: I see that we are adding twice for every message sent me: nope. i think that's important. Suraj: sorry for mentioning it after 3 slices me: I also checking RC3 why it slower than Hama 0.4. NP :-) Suraj: oh is the FT on? I knew it was going to degrade the performance. me: nope. looks like related with CompressionUtil Suraj: ok cool ... I might need more time with the confined recovery I am looking at the async messaging for now me: let's recreate RC. Suraj: ok Suraj: I have one more question on Combiners Suraj: we combine based on messages to one me: yes Suraj: but there is no way we can group these messages based on key and combine so say we are sending key value pair we can't combine .. we might need a new way of combiner along with what is there today me: i think user can do like that. Suraj: how? unless the messages are sorted or sent in a sorted way but we can return only one message My opinions might be biased to map-reduce implementation on bsp .... so :) me: http://wiki.apache.org/hama/Combiner if msg type is MapWritable<K, V>, not one. you can add multiple messages, Suraj: ok so you are proposing have a hashmap in a combiner? ok I got it
On Tue, Nov 6, 2012 at 4:52 PM, Edward J. Yoon <[email protected]> wrote: > Any other votes? > > Please feel free to vote. > > On Mon, Nov 5, 2012 at 11:55 PM, Tommaso Teofili > <[email protected]> wrote: >> I've found a couple of not trivial things to fix in GD BSP therefore I'd be >> happier to respin once I've committed the fixes (should be a couple of >> days), however I don't think it's worth blocking this (it's a 0.x in the >> end) so I vote +0. >> >> Regards, >> Tommaso >> >> >> 2012/11/5 Edward J. Yoon <[email protected]> >> >>> Works good for me +1 >>> >>> > BTW don't we want to add the gradient descent logistic and linear >>> > regression examples to the driver? We could also replace the normal PI >>> > Estimation example by the SuperstepPiEstimator. >>> >>> I'd like to add them to 0.7. >>> >>> On Sun, Nov 4, 2012 at 7:59 AM, Thomas Jungblut >>> <[email protected]> wrote: >>> > I have written a shellscript to download and check all the artifacts, >>> also >>> > updated the release check site in our committers guide. >>> > You can find it here: http://wiki.apache.org/hama/HowToCommit#line-42 >>> > If you mac guys have problems with executing the script, let me know via >>> > mail, we can change it to be compatible through all *nix systems. >>> > >>> > That said, the release checksums are fine. >>> > - maven clean install package works on source distribution without >>> problems >>> > - checked all examples with the localrunner and typical sample files >>> > >>> > BTW don't we want to add the gradient descent logistic and linear >>> > regression examples to the driver? We could also replace the normal PI >>> > Estimation example by the SuperstepPiEstimator. >>> > >>> > But still, I'm +1 (binding) >>> > >>> > 2012/11/2 Edward J. Yoon <[email protected]> >>> > >>> >> Hi all, >>> >> >>> >> Here's RC3: >>> >> http://people.apache.org/~edwardyoon/dist/0.6-RC3/ >>> >> >>> >> Tags: >>> >> http://svn.apache.org/repos/asf/hama/tags/0.6-RC3/ >>> >> >>> >> This RC fixes small bugs of unit tests. Please vote! >>> >> >>> >> Thanks. >>> >> -- >>> >> Best Regards, Edward J. Yoon >>> >> @eddieyoon >>> >> >>> >>> >>> >>> -- >>> Best Regards, Edward J. Yoon >>> @eddieyoon >>> > > > > -- > Best Regards, Edward J. Yoon > @eddieyoon -- Best Regards, Edward J. Yoon @eddieyoon
