Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Hive >> mail # user >> Re: Is my Use Case possible with Hive?

Copy link to this message
Re: Is my Use Case possible with Hive?
it is definitely possible to increase your performance.

I have run queries where more than 10 billion records were involved.
If you are doing joins in your queries, you may have a look at different
kind of joins supported by hive.
If one of your table is very small in size compared to another table then
you may consider mapside join etc

Also the number of maps and reducers are decided by the split size you
provide to maps.

I would suggest before you go full speed, decide on how you want to layout
data for hive.

You can try loading some data, partition the data and write queries based
on partition then performance will improve but in that case your queries
will be in batch processing format. there are other approaches as well.
On Mon, May 14, 2012 at 2:31 PM, Bhavesh Shah <[EMAIL PROTECTED]>wrote:

> That I fail to know, how many maps and reducers are there. Because due to
> some reason my instance get terminated   :(
> I want to know one thing that If we use multiple nodes, then what should
> be the count of maps and reducers.
> Actually I am confused about that. How to decide it?
> Also I want to try the different properties like block size, compress
> output, size of in-memorybuffer, parallel execution etc.
> Will these all properties matters to increase the performance?
> Nitin, you have read all my use case. Whatever the thing I did to
> implement with the help of Hadoop is correct?
> Is it possible to increase the performance?
> Thanks Nitin for your reply.   :)
> --
> Regards,
> Bhavesh Shah
> On Mon, May 14, 2012 at 2:07 PM, Nitin Pawar <[EMAIL PROTECTED]>wrote:
>> with a 10 node cluster the performance should improve.
>> how many maps and reducers are being launched?
>> On Mon, May 14, 2012 at 1:18 PM, Bhavesh Shah <[EMAIL PROTECTED]>wrote:
>>> I have near about 1 billion records in my relational database.
>>> Currently locally I am using just one cluster. But I also tried this on
>>> Amazon Elastic Mapreduce with 10 nodes. But the time taken to execute the
>>> complete program is same as that on my  single local machine.
>>> On Mon, May 14, 2012 at 1:13 PM, Nitin Pawar <[EMAIL PROTECTED]>wrote:
>>>> how many # records?
>>>> what is your hadoop cluster setup? how many nodes?
>>>> if you are running hadoop on a single node setup with normal desktop, i
>>>> doubt it will be of any help.
>>>> You need a stronger cluster setup for better query runtimes and
>>>> ofcourse query optimization which I guess you would have already taken care.
>>>> On Mon, May 14, 2012 at 12:39 PM, Bhavesh Shah <[EMAIL PROTECTED]
>>>> > wrote:
>>>>> Hello all,
>>>>> My Use Case is:
>>>>> 1) I have a relational database which has a very large data. (MS SQL
>>>>> Server)
>>>>> 2) I want to do analysis on these huge data  and want to generate
>>>>> reports
>>>>> on it after analysis.
>>>>> Like this I have to generate various reports based on different
>>>>> analysis.
>>>>> I tried to implement this using Hive. What I did is:
>>>>> 1) I imported all tables in Hive from MS SQL Server using SQOOP.
>>>>> 2) I wrote many queries in Hive which is executing using JDBC on Hive
>>>>> Thrift Server
>>>>> 3) I am getting the correct result in table form, which I am expecting
>>>>> 4) But the problem is that the time which require to execute is too
>>>>> much
>>>>> long.
>>>>>    (My complete program is executing in near about 3-4 hours on *small
>>>>> amount of data*).
>>>>>    I decided to do this using Hive.
>>>>>     And as I told previously how much time Hive consumed for
>>>>> execution. my
>>>>> organization is expecting to complete this task in near about less than
>>>>> 1/2 hours
>>>>> Now after spending too much time for complete execution for this task
>>>>> what
>>>>> should I do?
>>>>> I want to ask one thing that:
>>>>> *Is this Use Case is possible with Hive?* If possible what should I do
>>>>> in
>>>>> my program to increase the performance?
Nitin Pawar