+1, localOrShuffle will be a winner, as long as it's evenly distributing
work. If 1 tuple could say produce a variable 1-100 resultant tuples (and
these results were expensive enough to process, e.g. IO), it might well be
worth shuffling vs. localShuffling.

Michael Rose (@Xorlev <https://twitter.com/xorlev>)
Senior Platform Engineer, FullContact <http://www.fullcontact.com/>
[email protected]


On Wed, Mar 5, 2014 at 11:19 AM, Nathan Leung <[email protected]> wrote:

> In my experience on a 1 Gb network localOrShuffleGrouping was a clear
> winner in terms of performance.  But I haven't tested with 10 Gb, and if
> you have substantial business logic then that becomes a bigger factor than
> serializing/transferring data on the network.  I think the performance of
> any given grouping is too dependent on your business logic; it will be
> difficult to quantify how well it performs in a canned benchmark.  And
> sometimes your business logic will define a grouping for you (e.g. fields
> grouping) whether it's the best performer or not.
>
>
> On Wed, Mar 5, 2014 at 1:05 PM, Roberto Coluccio <
> [email protected]> wrote:
>
>> Hello Michael, thanks for your feedback.
>>
>> I'm looking for a performance comparison. I know that not all the
>> policies are "really comparable", but even obvious comparisons all listed
>> together could be a useful reference.
>>
>> Roberto
>>
>>
>> On Wed, Mar 5, 2014 at 6:58 PM, Michael Rose <[email protected]>wrote:
>>
>>> What kind of comparisons are you looking for? How they functionally work?
>>>
>>> Michael Rose (@Xorlev <https://twitter.com/xorlev>)
>>> Senior Platform Engineer, FullContact <http://www.fullcontact.com/>
>>> [email protected]
>>>
>>>
>>> On Wed, Mar 5, 2014 at 9:52 AM, Roberto Coluccio <
>>> [email protected]> wrote:
>>>
>>>> Hello folks,
>>>>
>>>> I was unable to find any complete example (or, better, related work in
>>>> the scientific literature) in which (almost) all the *stream grouping
>>>> policies* have been used and compared. Do you have any reference you
>>>> could please share with me?
>>>>
>>>> Thank you and best regards,
>>>>
>>>> Roberto Coluccio
>>>>
>>>
>>>
>>
>

Reply via email to