hadoop-hdfs-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ajay Srivastava <Ajay.Srivast...@guavus.com>
Subject Re: io.sort.factor
Date Wed, 23 Jan 2013 07:16:06 GMT
Hi Bharat,

I am looking at these logs -
2013-01-22 07:35:42,923 INFO org.apache.hadoop.mapred.MapTask: Finished spill 2

The number at the end of string does not go beyond 6. So I assume you are correct.


Regards,
Ajay Srivastava


On 23-Jan-2013, at 12:14 PM, bharath vissapragada wrote:

Hi,

>From my understanding, increasing "io.sort.mb" decreases the number of disk flushes as
more data is spilled at once and this boosts the performance. This obviously decreases the
number of spills.

One possibility in this case is that, the number of spills have become less than "io.sort.factor"
and all the spills are getting merged in a single go (per partition). So even if you increase
the "io.sort.factor" value now, it shouldn't make any difference. So check the logs and see
the no. of spills created and if its less than "io.sort.factor".

This is a wild guess, please feel free to correct me if I'm wrong.

Thanks



On Wed, Jan 23, 2013 at 11:53 AM, Ajay Srivastava <Ajay.Srivastava@guavus.com<mailto:Ajay.Srivastava@guavus.com>>
wrote:
Hi,

io.sort.factor  --  The number of streams to merge at once while sorting files. This determines
the number of open file handles.


How can I use this parameter to improve performance of mapreduce job?
My understanding from above description was If there are many spill records then increasing
io.sort.mb as well as io.sort.factor will help in better performance. Increasing io.sort.mb
helped but changing io.sort.factor (> 10) does not seem to improve/degrade performance
of mapred  job.



Regards,
Ajay Srivastava



Mime
View raw message