You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@hive.apache.org by Erwan MAS <er...@mas.nom.fr> on 2015/04/02 18:01:53 UTC

Hive and engine performance tez vs mr

Hello ,

I have a issue on hive , with tez engine . When  try to execute a query , with 
tez engine , the query is 9 times slower than map/reduce . 

The query is a left outer join on two table using orc storage .

With map/reduce i have  :
Job 0 : Map 27 Reduce 256
Job 1 : Map 27 Reduce 256
Time taken 110 sec

With tez i have :
Map 1 :  1/1 Map 4 : 3/3 Reducer 2: 256/256 Reducer 3: 256/256
Time taken 930 sec 

With my configuration tez want to use only one mapper for some part .

How to increase this number of mapper ? 
Which variable on hive , i must set to change this behavior  ?

My context :
   Hive 0.13 on Hortonworks 2.1

--
     ____________________________________________________________
    / Erwan MAS                                                 /\
   | mailto:erwan@mas.nom.fr                                   |_/
___|________________________________________________________   |
\___________________________________________________________\__/

Re: Hive and engine performance tez vs mr

Posted by Carter Shanklin <ca...@hortonworks.com>.
Erwan,

Faced with a similar situation last week I found that decreasing

mapred.max.split.size

Increased my parallelism by 6x. Yes mapred even though it was a Tez job. I
reduced it to 10mb from 256mb which I believe is the default.

The other variables to try are:
tez.grouping.min-size (make it smaller)
tez.grouping.max-size (smaller as well)


Good luck.


On 4/6/15, 2:57 PM, "Erwan MAS" <er...@mas.nom.fr> wrote:

>On Mon, Apr 06, 2015 at 12:15:05PM -0500, max scalf wrote:
>> Try setting the below in Hive and see what happens..btw what are you
>> configs in hive if any?
>> 
>> set mapred.map.tasks = 20;
>> 
>
>Does not change the behavior :(
>
>--
>     ____________________________________________________________
>    / Erwan MAS                                                 /\
>   | mailto:erwan@mas.nom.fr                                   |_/
>___|________________________________________________________   |
>\___________________________________________________________\__/


Re: Hive and engine performance tez vs mr

Posted by Erwan MAS <er...@mas.nom.fr>.
On Mon, Apr 06, 2015 at 12:15:05PM -0500, max scalf wrote:
> Try setting the below in Hive and see what happens..btw what are you
> configs in hive if any?
> 
> set mapred.map.tasks = 20;
> 

Does not change the behavior :(

--
     ____________________________________________________________
    / Erwan MAS                                                 /\
   | mailto:erwan@mas.nom.fr                                   |_/
___|________________________________________________________   |
\___________________________________________________________\__/

Re: Hive and engine performance tez vs mr

Posted by max scalf <or...@gmail.com>.
Try setting the below in Hive and see what happens..btw what are you
configs in hive if any?

set mapred.map.tasks = 20;


On Thu, Apr 2, 2015 at 11:01 AM, Erwan MAS <er...@mas.nom.fr> wrote:

> Hello ,
>
> I have a issue on hive , with tez engine . When  try to execute a query ,
> with
> tez engine , the query is 9 times slower than map/reduce .
>
> The query is a left outer join on two table using orc storage .
>
> With map/reduce i have  :
> Job 0 : Map 27 Reduce 256
> Job 1 : Map 27 Reduce 256
> Time taken 110 sec
>
> With tez i have :
> Map 1 :  1/1 Map 4 : 3/3 Reducer 2: 256/256 Reducer 3: 256/256
> Time taken 930 sec
>
> With my configuration tez want to use only one mapper for some part .
>
> How to increase this number of mapper ?
> Which variable on hive , i must set to change this behavior  ?
>
> My context :
>    Hive 0.13 on Hortonworks 2.1
>
> --
>      ____________________________________________________________
>     / Erwan MAS                                                 /\
>    | mailto:erwan@mas.nom.fr                                   |_/
> ___|________________________________________________________   |
> \___________________________________________________________\__/
>