Questions on hadoop configuration for heterogeneous cluster

[Yu Li <ca...@gmail.com>]

Hi all,

I'm a little confused about how to configure hadoop in a heterogeneous
cluster. For example, if I have one machine(m1) with a two-core
processor, another(m2) with a four-core processor, and I'd like to use them
as tasktracker nodes in a hadoop cluster, how could I configure the
mapred.tasktracker.map/reduce.tasks.maximum? Could I set both parameters to
2 on m1, and set to 4 on m2? Or I have to set both to 2 on JT node? In
another word, among those massive of parameters in *-site.xml and
environtment variables in hadoop-env.sh, which ones could be set on each
DN/TT with different values and still take effect?

Thanks in advance and look forward to your reply.

-- 
Best Regards,
Li Yu

Re: Questions on hadoop configuration for heterogeneous cluster

[Yu Li <ca...@gmail.com>]

Hi Harsh,

Thanks a lot for your reply, this really helps!

On 16 December 2010 18:08, Harsh J <qw...@gmail.com> wrote:

> Hi,
>
> On Thu, Dec 16, 2010 at 2:48 PM, Yu Li <ca...@gmail.com> wrote:
> > Hi all,
> >
> > I'm a little confused about how to configure hadoop in a heterogeneous
> > cluster. For example, if I have one machine(m1) with a two-core
> > processor, another(m2) with a four-core processor, and I'd like to use
> them
> > as tasktracker nodes in a hadoop cluster, how could I configure the
> > mapred.tasktracker.map/reduce.tasks.maximum? Could I set both parameters
> to
> > 2 on m1, and set to 4 on m2?
>
> Yes you can do this just fine. Note that the configuration property
> name says "tasktracker" meaning it is a tasktracker specific setting
> and can vary for each. Has nothing to do with the JobTracker.
>
> > another word, among those massive of parameters in *-site.xml and
> > environtment variables in hadoop-env.sh, which ones could be set on each
> > DN/TT with different values and still take effect?
>
> *.tasktracker.* and *.datanode.* properties are TT and DN specific and
> can be set individually for each of them. This is due to a naming
> convention followed by Hadoop.
>
> --
> Harsh J
> www.harshj.com
>



-- 
Best Regards,
Li Yu

Re: Questions on hadoop configuration for heterogeneous cluster

[Harsh J <qw...@gmail.com>]

Hi,

On Thu, Dec 16, 2010 at 2:48 PM, Yu Li <ca...@gmail.com> wrote:
> Hi all,
>
> I'm a little confused about how to configure hadoop in a heterogeneous
> cluster. For example, if I have one machine(m1) with a two-core
> processor, another(m2) with a four-core processor, and I'd like to use them
> as tasktracker nodes in a hadoop cluster, how could I configure the
> mapred.tasktracker.map/reduce.tasks.maximum? Could I set both parameters to
> 2 on m1, and set to 4 on m2?

Yes you can do this just fine. Note that the configuration property
name says "tasktracker" meaning it is a tasktracker specific setting
and can vary for each. Has nothing to do with the JobTracker.

> another word, among those massive of parameters in *-site.xml and
> environtment variables in hadoop-env.sh, which ones could be set on each
> DN/TT with different values and still take effect?

*.tasktracker.* and *.datanode.* properties are TT and DN specific and
can be set individually for each of them. This is due to a naming
convention followed by Hadoop.

-- 
Harsh J
www.harshj.com