Hi Bhavesh I'm not sure of AWS, but from a quick reading cluster wide settings like hdfs block size can be set on hdfs-site.xml through bootstrap actions. Since you are changing hdfs block size set min and max split size across the cluster using bootstrap actions itself. The rest of the properties can on set on a per job level.
Doesn't AWS provide an option to use "hive -f"? If so, just provide all the properties required for tuning the query followed by queries(in order) in a file and simply execute it using "hive -f <file name>". Regards Bejoy KS ________________________________ From: Bhavesh Shah <bhavesh25s...@gmail.com> To: user@hive.apache.org; Bejoy Ks <bejoy...@yahoo.com> Sent: Tuesday, May 8, 2012 3:33 PM Subject: Re: Want to improve the performance for execution of Hive Jobs. Thanks Bejoy KS for your reply, I want to ask one thing that If I want to set this parameter on Amazon Elastic Mapreduce then how can I set these variable like: e.g. SET mapred.min.split.size=m; SET mapred.max.split.size=m+n; set dfs.block.size=128 set mapred.compress.map.output=true set io.sort.mb=400 etc.... For all this do I need to write shell script for setting this variables on the particular path /home/hadoop/hive/bin/hive -e 'set .....' or pass all this steps in bootstrap actions??? I found this link to pass the bootstrap actions http://docs.amazonwebservices.com/ElasticMapReduce/latest/DeveloperGuide/Bootstrap.html#BootstrapPredefined What should I do in such case?? On Tue, May 8, 2012 at 2:55 PM, Bejoy Ks <bejoy...@yahoo.com> wrote: Hi Bhavesh > > > In sqoop you can optimize the performance by using --direct mode for >import and increasing the number of mappers used for import. When you increase >the number of mappers you need to ensure that the RDBMS connection pool will >handle those number of connections gracefully. Also use a evenly distributed >column as --split-by, that'll ensure that all mappers are kind of equally >loaded. > min split size and map split size can be set on a job level. But, there are >chances of slight loss in data locality if you increase these values. By >increasing these values you are increasing the data volume processed per >mapper so less number of mappers , now you need to see whether this will that >get you substantial performance gains. I havent seen much gains there when I >tried out those on some of my workflows in the past. A better approach than >this would be increasing the hdfs block size itself if your cluster deals with >relatively larger files. Of you change the hdfs block size then make the >changes accordingly on min split and max split values. > You can set all min and max split sizes using SET command in hive CLI >itself. >hive> SET mapred.min.split.size=m; >hive> SET mapred.max.split.size=m+n; > > >Regards >Bejoy KS > > > > >________________________________ > From: Bhavesh Shah <bhavesh25s...@gmail.com> >To: user@hive.apache.org >Sent: Tuesday, May 8, 2012 11:35 AM >Subject: Re: Want to improve the performance for execution of Hive Jobs. > > > >Thanks Both of you for their replies, >If I decide to deploy my JAR on Amazon Elastic Mapreduce then, > >1) Default block size is 64 MB, so insuch case I have to set it to 128 MB..... >is it right??? >2) Amazon EMR has already values for mapred.min.split.size >and mapred.max.split.size, and mapper and reducer too. So is there any need to >set the values there? If yes then how to set for all clusters? Is it possible >by setting all these above parameters in --bootstrap-actions.... to apply this >for all nodes while submitting jobs to Amazon EMR?? > >Thanks both of u very much > >-- >Regards, >Bhavesh Shah > > >On Tue, May 8, 2012 at 11:19 AM, Mapred Learn <mapred.le...@gmail.com> wrote: > >Try setting this value to your block >>Size, for 128 mb block size, >> >> >>set mapred.min.split.size=128000 >>Sent from my iPhone >> >>On May 7, 2012, at 10:11 PM, Bhavesh Shah <bhavesh25s...@gmail.com> wrote: >> >> >>Thanks Nitin for your reply. >>> >>>In short my Task is >>>1) Initially I want to import the data from MS SQL Server into HDFS using >>>SQOOP. >>>2) Through Hive I am processing the data and generating the result in one >>>table >>>3) That result containing table from Hive is again exported to MS SQL SERVER >>>back. >>> >>>Actually the data which I am importing from MS SQL Server is very large (near about 5,00,000 entries in one table. Like wise I have 30 tables). For this I have written a task in Hive which contains only queries (And each query has used a lot of joins in it). So due to this the performance is very poor on my single local machine ( It takes near about 3 hrs to execute completely). I have observed that when I have submitted a single query to Hive CLI it took 10-11 jobs to execute completely. >>> >>>set mapred.min.split.size >>>set mapred.max.split.size >>>Should this value to be set in bootstrap action while submitting jobs to >>>amazon EMR? What value to be set for it as I don't know? >>> >>> >>>-- >>>Regards, >>>Bhavesh Shah >>> >>> >>>On Tue, May 8, 2012 at 10:31 AM, Nitin Pawar <nitinpawar...@gmail.com> wrote: >>> >>>1) check the jobtracker url to see how many maps/reducers have been launched >>>>2) if you have a large dataset and wants to execute it fast, you >>>>set mapred.min.split.size and mapred.max.split.size to an optimal value so >>>>that more mappers will be launched and will finish >>>>3) if you are doing joins, there are different ways to go according to the >>>>data you have and size of data >>>> >>>> >>>>it will be helpful if you can let us know your datasizes and query details >>>> >>>> >>>> >>>>On Tue, May 8, 2012 at 10:07 AM, Bhavesh Shah <bhavesh25s...@gmail.com> >>>>wrote: >>>> >>>>Hello all, >>>>>I have written a Hive JDBC code and created a JAR of it. I am running that >>>>>JAR on 10 cluster. >>>>>But the problem as I am using the 10 cluster still the performance is same >>>>>as that on single cluster. >>>>> >>>>>What to do to improve the performance of Hive Jobs? Is there anything >>>>>configuration setting to set before the submitting Hive Jobs to cluster? >>>>>One more thing I want to know is that How can we come to know that is job >>>>>running on all cluster? >>>>> >>>>>Please let me know if anyone knows about it? >>>>> >>>>>-- >>>>>Regards, >>>>>Bhavesh Shah >>>>> >>>> >>>> >>>> >>>>-- >>>>Nitin Pawar >>>> >>>> >>> >>> > > > > > -- Regards, Bhavesh Shah
