Re: Issues with performance on Hadoop/Hive

[Brian Bockelman]

On Sep 3, 2009, at 11:53 PM, Ramiya V wrote:

Hi,

Thanks Amandeep and Ashish!

@Ashish: I have set the hive.metastore.warehouse.dir parameter as / home/hive/warehouse. This warehouse directory is on the local filesystem. So will the tables now get stored on the local filesystem or HDFS? I mean do we need to specify explicitly if we want the path to refer a location on the local filesystem? @Amandeep: Actually I need to know how the data gets distributed across the cluster. The master machine has 70GB free space and the 3 slaves have 140GB,50GB,140GB free space on Ubuntu. So when I load a table on Hive with 45GB of data,how will it get distributed across 4 nodes? I mean since master has 70GB free space will it store all the data on the master itself? (This I observed as when I loaded the table, 45GB data was stored on master and some blocks were replicated on the other 3 slaves) I have set dfs.replication factor as 2. I wanted to know how exactly Hadoop uses its intelligence to utilize the free space effectively to store data on HDFS on a cluster,

Hey Ramya,



http://hadoop.apache.org/common/docs/current/hdfs_design.html

Brian

I believe this is thoroughly described in the system architecture document. Let us know if there's something you feel is missing:

-Ramya
________________________________________
From: Ashish Thusoo [athu...@facebook.com]
Sent: Wednesday, September 02, 2009 11:02 PM
To: common-user@hadoop.apache.org
Subject: RE: Issues with performance on Hadoop/Hive

Hi Ramya,



/user/hive/warehouse in hdfs.

Ashish

-----Original Message-----
From: Amandeep Khurana [mailto:ama...@gmail.com]
Sent: Wednesday, September 02, 2009 12:52 AM
To: common-user@hadoop.apache.org
Subject: Re: Issues with performance on Hadoop/Hive

Answers inline


Amandeep Khurana
Computer Science Graduate Student
University of California, Santa Cruz

If you are using the hive-default.xml and have not overwritten the hive.metastore.warehouse.dir parameter then by default the tables will get placed in On Tue, Sep 1, 2009 at 10:08 PM, Ramiya V <ramiy...@persistent.co.in> wrote:

Hi,

I have set up a 4 (physica) nodes Hadoop cluster. Configuration: 2GB
RAM each machine. Currently am using the sub-project Hive for firing
queries on 45GB of data. I have certain queries that need to be
resolved:-

1) The performance that I am getting with the above setup is quite
bad. It takes app 39 minutes for simple select query (with where

Is this setting good enough for the above setup? Are there any

clause). I have set the mapred.map.tasks=13 and mapred.reduce.tasks=7. significant configuration parameters I need to set for getting a better performance on Hive?

Check on the resource utilization. I think you shouldnt be running more than 3 mappers + 1 reducer on each node at any time (given the hardware you are using). But then that mostly depends on the amount of work being done in the mappers and reducers.

2) Does anybody know how exactly the data on HDFS is distributed across

nodes in a cluster? Also when we load the tables in Hive (by firing
Load command on master node),how and where is the data placed on HDFS
in a cluster?

Files are divided into blocks and the blocks are stored on the Datanodes.. Each block is 64MB by default. I'm not sure how the blocks are distributed among the datanodes.. 3) How and when does the data replication for HDFS take place in a cluster?

Currently I have set the dfs replication factor=1. How does this
affect the performance?

Once you put the data into the hdfs, it starts replicating the blocks.
However, the put is successful as soon as one block gets created...

3) Does adding a Virtual Machine to a physical machine cluster bring
about significant degradation in the performance?

Dont have numbers for this, but it does impact the performance. Moreover, your hardware resources are low and there is really no value add in using virtual machines on top of it.

Please let me know asap.

Thanks,
Ramya




DISCLAIMER
==========
This e-mail may contain privileged and confidential information which

use of the individual or entity to which it is addressed. If you are
not the intended recipient, you are not authorized to read, retain,

Persistent Systems Ltd. does not accept any liability for virus
infected mails.

is the property of Persistent Systems Ltd. It is intended only for the copy, print, distribute or use this message. If you have received this communication in error, please notify the sender and delete all copies of this message.

DISCLAIMER
==========

This e-mail may contain privileged and confidential information which is the property of Persistent Systems Ltd. It is intended only for the use of the individual or entity to which it is addressed. If you are not the intended recipient, you are not authorized to read, retain, copy, print, distribute or use this message. If you have received this communication in error, please notify the sender and delete all copies of this message. Persistent Systems Ltd. does not accept any liability for virus infected mails.

smime.p7s
Description: S/MIME cryptographic signature