hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sujee Maniyam <su...@sujee.net>
Subject Re: Several questions about running HBase on EC2
Date Thu, 22 Apr 2010 05:55:24 GMT
Nice presentation Andy!

Sean,
I am experimenting with a small cluster on EC2 right now.  Here is my
experience.

1) it is a 5 node cluster (1 master + 4 slaves).  All c1.xlarge instances.

2) I initially tried m1.large, but ran into some stability issues.  So
moved to c1.xlarge.  Cluster is more stable now.
Follow the instructions here carefully :
http://hadoop.apache.org/hbase/docs/r0.20.3/api/overview-summary.html

3) I made a custom AMI from Ubuntu v9.10.  It had all the required
software.  I spin up the instances based on this AMI.
If I do it again, I'd go with Cloudera AMIs

4) I setup the cluster manually; it was manageable b/c it is a small
cluster.  If you need more machines, look at the EC2 scripts in Andy's
presentation.

5) I run the whole cluster on EBS volumes.  The max limit on EBS is
1TB.  You can always string a few EBS disks as RAID-X configuration

6) EBS through-put is NOT really that impressive.  Understandably.
But good-enough

6.5) format your volumes as EXT4 or XFS file system.  Both of these
are better than EXT3.

7) turn on compression ( I use LZO).  Not only it saves space on HDFS,
it also speeds up read/writes.

8) c1.xlarge has 8G mem.  My Hbase heap is 3G.

9) I have about 600 million rows in hbase (1 TB compressed) .  I run
MR jobs on them.
I am fighting some stability problems now.  Probably because the
cluster is too small to handle the load (2000 regions / regionserver).
 Investigating and will post results here.

10) part of the experimentation is to figure out the cost of running a
cluster as well.   a C1.Xlarge costs about $500 / month - reserved
instances cost about $150 / month (1 yr term).
So be sure to factor this as well.

11) hbase comes with simple load-tester.
(HBASE/bin/hbase    org.apache.hadoop.hbase.PerformanceEvaluation)
useful to test the cluster and it's stability.


hope this helps.

thanks
Sujee
http://sujee.net
hbase-map-reduce tutorial here :
http://sujee.net/tech/articles/hbase-map-reduce-freq-counter/




On Wed, Apr 21, 2010 at 12:40 AM, Andrew Purtell <apurtell@apache.org> wrote:
> More info:
>
>  http://hbase.s3.amazonaws.com/hbase/HBase-EC2-HUG9.odp
>  http://hbase.s3.amazonaws.com/hbase/HBase-EC2-HUG9.pdf
>
>  - Andy
>
>
>
>
>
>

Mime
View raw message