Hi users,

I am working on a CDH3 cluster of 12 nodes (Task Trackers running on all
the 12 nodes and 1 node running the Job Tracker).
In order to perform a WordCount benchmark test, I did the following:

   - Executed "RandomTextWriter" first to create 100 GB data (Note that I
   have changed the "test.randomtextwrite.total_bytes" parameter only, rest
   all are kept default).
   - Next, executed the "WordCount" program for that 100 GB dataset.

The "Block Size" in "hdfs-site.xml" is set as 128 MB. Now, according to my
calculation, total number of Maps to be executed by the wordcount job
should be 100 GB / 128 MB or 102400 MB / 128 MB = 800.
But when I am executing the job, it is running a total number of 900 Maps,
i.e., 100 extra.
So, why this extra number of Maps? Although, my job is completing
successfully without any error.

Again, if I don't execute the "RandomTextWwriter" job to create data for my
wordcount, rather I put my own 100 GB text file in HDFS and run
"WordCount", I can then see the number of Maps are equivalent to my
calculation, i.e., 800.

Can anyone tell me why this odd behaviour of Hadoop regarding the number of
Maps for WordCount only when the dataset is generated by RandomTextWriter?
And what is the purpose of these extra number of Maps?

Regards,
Gaurav Dasgupta

Hi Gaurav,

Did you turn off speculative execution?

Best Regards,
Anil

On Aug 16, 2012, at 7:13 AM, Gaurav Dasgupta <[EMAIL PROTECTED]> wrote:

> Hi users,
>  
> I am working on a CDH3 cluster of 12 nodes (Task Trackers running on all the 12 nodes and 1 node running the Job Tracker).
> In order to perform a WordCount benchmark test, I did the following:
> Executed "RandomTextWriter" first to create 100 GB data (Note that I have changed the "test.randomtextwrite.total_bytes" parameter only, rest all are kept default).
> Next, executed the "WordCount" program for that 100 GB dataset.
> The "Block Size" in "hdfs-site.xml" is set as 128 MB. Now, according to my calculation, total number of Maps to be executed by the wordcount job should be 100 GB / 128 MB or 102400 MB / 128 MB = 800.
> But when I am executing the job, it is running a total number of 900 Maps, i.e., 100 extra.
> So, why this extra number of Maps? Although, my job is completing successfully without any error.
>  
> Again, if I don't execute the "RandomTextWwriter" job to create data for my wordcount, rather I put my own 100 GB text file in HDFS and run "WordCount", I can then see the number of Maps are equivalent to my calculation, i.e., 800.
>  
> Can anyone tell me why this odd behaviour of Hadoop regarding the number of Maps for WordCount only when the dataset is generated by RandomTextWriter? And what is the purpose of these extra number of Maps?
>  
> Regards,
> Gaurav Dasgupta

Hi Anil,

The speculative execution property is off from the begining.
In addition to my previous mail, I would like to add some more points:

I have checked the same while generating 100 GB data from RandomTextWriter,
where "hadoop fs -dus <hdfs output dir>" gives me 102.65 GB.
So if I calculate the number of Maps while running WordCount on this should
be ((102.65 * 1024) MB / 128 MB) = 821.22. So, there should be 822 Maps
running, but the actual number of Maps running are 900, i.e., extra 78.
Also, the number of Maps ran for the above RandomTextWriter were 100.

Above is not the same if I generate data using TeraGen (hadoop jar
hadoop-examples.jar teragen 10000 <output_dir>) and then perform WordCount
on it, it gives me the number of Maps = 2 (Note: Number of Maps for TeraGen
was also 2).
Please find in the attachement the screenshots of the JobTracker UI for
RandomTextWriter and WordCount for your reference.

Regards,
Gaurav Dasgupta
On Thu, Aug 16, 2012 at 7:57 PM, Anil Gupta <[EMAIL PROTECTED]> wrote:

>  Hi Gaurav,
>
> Did you turn off speculative execution?
>
> Best Regards,
> Anil
>
> On Aug 16, 2012, at 7:13 AM, Gaurav Dasgupta <[EMAIL PROTECTED]> wrote:
>
>   Hi users,
>
> I am working on a CDH3 cluster of 12 nodes (Task Trackers running on all
> the 12 nodes and 1 node running the Job Tracker).
> In order to perform a WordCount benchmark test, I did the following:
>
>    - Executed "RandomTextWriter" first to create 100 GB data (Note that I
>    have changed the "test.randomtextwrite.total_bytes" parameter only, rest
>    all are kept default).
>    - Next, executed the "WordCount" program for that 100 GB dataset.
>
> The "Block Size" in "hdfs-site.xml" is set as 128 MB. Now, according to my
> calculation, total number of Maps to be executed by the wordcount job
> should be 100 GB / 128 MB or 102400 MB / 128 MB = 800.
> But when I am executing the job, it is running a total number of 900 Maps,
> i.e., 100 extra.
> So, why this extra number of Maps? Although, my job is completing
> successfully without any error.
>
> Again, if I don't execute the "RandomTextWwriter" job to create data for
> my wordcount, rather I put my own 100 GB text file in HDFS and run
> "WordCount", I can then see the number of Maps are equivalent to my
> calculation, i.e., 800.
>
> Can anyone tell me why this odd behaviour of Hadoop regarding the number
> of Maps for WordCount only when the dataset is generated by
> RandomTextWriter? And what is the purpose of these extra number of Maps?
>
> Regards,
> Gaurav Dasgupta
>
>

Hi Gaurav

How many input files are there for the wordcount map reduce job? Do you
have input files lesser than a block size? If you are using the default
TextInputFormat there will be one task generated per file for sure, so if
you have  files less than block size the calculation specified here for
number of splits won't hold. If small files are there then definitely the
number of maps tasks should be more.

Also did you change the split sizes as well along with block size?

Regards
Bejoy KS

Hi Gaurav

To add on more clarity to my previous mail
If you are using the default TextInputFormat there will be *atleast* one
task generated per file even if the file size is less than
the block size. (assuming you have split size equal to block size)

So the right way to calculate the number of splits is per file and not on
the whole input data size. Calculate number of blocks per file and summing
up those values from all files would equate to the number of mappers.

What is the value of mapred.max.splitsize in your job? If it is less than
the hdfs block size there will be more spits for even for a hdfs block.

Regards
Bejoy KS

Hi Bejoy,

The total number of Maps in the RandomTextWriter execution were 100 and
hence the total number of input files for WordCount are 100.
My dfs.block.size = 128MB and I have not changed the
mapred.max.split.size and could not find it in myJob.xml file.
Hence refering the formula *max(minsplitsize, min(maxsplitsize, blocksize))*,
I am assuming the mapred.max.split.size to be 128MB.
If I calculate the blocks per file [bytes per file / block size (128 MB)]
gives me 8.21 for all. And then if I sum up them it becomes 821.22 (Same as
my previous calculation).

I have some how managed to do a need copy of the Job.xml in a word doc. I
copied it from browser as I cannot recover it in the hdfs. Please find it
in the attachment. You may refer the parameters and configuration there. I
have also attached the console output for the bytes per file in the
WordCount input.

Regards,
Gaurav Dasgupta
On Fri, Aug 17, 2012 at 3:28 PM, Bejoy Ks <[EMAIL PROTECTED]> wrote:

> Hi Gaurav
>
> To add on more clarity to my previous mail
> If you are using the default TextInputFormat there will be *atleast* one
> task generated per file even if the file size is less than
> the block size. (assuming you have split size equal to block size)
>
> So the right way to calculate the number of splits is per file and not on
> the whole input data size. Calculate number of blocks per file and summing
> up those values from all files would equate to the number of mappers.
>
> What is the value of mapred.max.splitsize in your job? If it is less than
> the hdfs block size there will be more spits for even for a hdfs block.
>
> Regards
> Bejoy KS
>
>
>
>

Hi Gaurav

While calculating you got the number of map tasks per file as 8.12 ie 9 map
tasks for each file. So for 100 files it is 900 map tasks and now your
numbers match. Doesn't it look right?

Regards
Bejoy KS

Hi

I have got it. It was my mistake understanding the calculation. Thanks for
the help.

Regards,
Gaurav Dasgupta

On Fri, Aug 17, 2012 at 5:15 PM, Bejoy Ks <[EMAIL PROTECTED]> wrote:

> Hi Gaurav
>
> While calculating you got the number of map tasks per file as 8.12 ie 9
> map tasks for each file. So for 100 files it is 900 map tasks and now your
> numbers match. Doesn't it look right?
>
> Regards
> Bejoy KS
>