Hi all

I'm doing a test and need create lots of files ( 100 million ) in HDFS， I use a shell script to do this , it's very very slow, how to create a lot files in HDFS quickly?
Thanks

First, it is virtually impossible to create 100 million files in HDFS
because the name node can't hold that many.

Secondly, file creation is bottle-necked by the name node so the files that
you can create can't be created at more than about 1000 per second (and
achieving more than half that rate is somewhat difficult).

Thirdly, you need to check your cluster size because each data node can only
store a limited number of blocks (exactly how many differs from version to
version of Hadoop).  For small clusters this is a more exigent limit than
the size limit of the name node.

Why is it that you need to do this?

Perhaps there is a work-around?  Consider for instance HAR files:

http://www.cloudera.com/blog/2009/02/the-small-files-problem/
2011/5/29 ccxixicc <[EMAIL PROTECTED]>

> Hi all
> I'm doing a test and need create lots of files ( 100 million ) in HDFS��� I
> use a shell script to do this , it's very very slow, how to create a lot
> files in HDFS quickly?
> Thanks
>

Your best bet would be to take a look at synthetic load generator.

10^8 files would be a problem for most cases because you'd need to have a
really beefy NN for that (~48GB of JVM heap and all that). The biggest I've
heard about hold something at the order of 1.15*10^8 objects (files & dirs)
and is serving a largest Hadoop cluster in the world for Yahoo! production
setup. You might want to check YDN for more details about this case, I guess.

Hope it helps,
  Cos

On Mon, May 30, 2011 at 10:44AM, ccxixicc wrote:
>    Hi all
>    I'm doing a test and need create lots of files ( 100 million ) in
>    HDFS-L-NOT I use a shell script to do this , it's very very slow, how to
>    create a lot files in HDFS quickly?
>    Thanks

I don't know what your use case is, but you may want to investigate things like hbase or Cassandra or voldemort if you need lots of small files

---
Ian Holsman - 703 879-3128

I saw the angel in the marble and carved until I set him free -- Michelangelo

On 30/05/2011, at 12:54 AM, Konstantin Boudnik <[EMAIL PROTECTED]> wrote:

> Your best bet would be to take a look at synthetic load generator.
>
> 10^8 files would be a problem for most cases because you'd need to have a
> really beefy NN for that (~48GB of JVM heap and all that). The biggest I've
> heard about hold something at the order of 1.15*10^8 objects (files & dirs)
> and is serving a largest Hadoop cluster in the world for Yahoo! production
> setup. You might want to check YDN for more details about this case, I guess.
>
> Hope it helps,
>  Cos
>
> On Mon, May 30, 2011 at 10:44AM, ccxixicc wrote:
>>   Hi all
>>   I'm doing a test and need create lots of files ( 100 million ) in
>>   HDFS-L-NOT I use a shell script to do this , it's very very slow, how to
>>   create a lot files in HDFS quickly?
>>   Thanks