You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@hbase.apache.org by Sujee Maniyam <su...@sujee.net> on 2011/09/12 09:22:29 UTC
tutorial : HBase performance testing
Hi all
I have an article online, that shows how do some quick benchmarks on a HBase
cluster.
http://sujee.net/tech/articles/hbase-performance-testing/
feedbacks / thoughts welcome.
thanks very much
Sujee Maniyam
http://sujee.net
Re: tutorial : HBase performance testing
Posted by Stack <st...@duboce.net>.
nice. thanks.
St.Ack
On Mon, Sep 12, 2011 at 12:22 AM, Sujee Maniyam <su...@sujee.net> wrote:
> Hi all
> I have an article online, that shows how do some quick benchmarks on a HBase
> cluster.
> http://sujee.net/tech/articles/hbase-performance-testing/
>
> feedbacks / thoughts welcome.
>
> thanks very much
> Sujee Maniyam
> http://sujee.net
>
Re: tutorial : HBase performance testing
Posted by Jean-Daniel Cryans <jd...@apache.org>.
On Tue, Sep 13, 2011 at 11:20 PM, Sujee Maniyam <su...@sujee.net> wrote:
> hehe J-D (hopefully first name!)
:)
> I agree with your point that pre-splitting the table can make a big
> difference.
>
> Do the later versions of 'PerformanceEvaluation' class has an option to
> pre-split the table? I remember, when I ran this for the first time,
> only one region server is busy until the table split. But second time
> around, all the region-servers were hit with requests.
No, this wasn't added. Nicolas had this idea tho:
https://issues.apache.org/jira/browse/HBASE-4163
> I just peeked at the code for this class, and it does NOT truncate the
> table. So subsequent runs benefit from split tables already. And it looks
> like it is overriding the rows.
>
> ** So I will mention this, and say to ignore the first run and only measure
> subsequent runs. what do you think? **
It will still be splitting a ton, even tho it's on multiple servers.
J-D
Re: tutorial : HBase performance testing
Posted by Sujee Maniyam <su...@sujee.net>.
hehe J-D (hopefully first name!)
I agree with your point that pre-splitting the table can make a big
difference.
Do the later versions of 'PerformanceEvaluation' class has an option to
pre-split the table? I remember, when I ran this for the first time,
only one region server is busy until the table split. But second time
around, all the region-servers were hit with requests.
I just peeked at the code for this class, and it does NOT truncate the
table. So subsequent runs benefit from split tables already. And it looks
like it is overriding the rows.
** So I will mention this, and say to ignore the first run and only measure
subsequent runs. what do you think? **
thanks for your feedback.
Sujee
http://sujee.net
On Mon, Sep 12, 2011 at 5:05 PM, Jean-Daniel Cryans <jd...@apache.org>wrote:
> Hi Sujee,
>
> Both tools are lacking a step where they create the tables pre-split.
> The difference can be staggering and possibly lead to misguiding.
>
> Also you referred to "George Lars", I don't understand those crazy
> Europeans that give their child two first names either (joking) but
> I'm 99.9999% sure it's the other way around :)
>
>
Re: tutorial : HBase performance testing
Posted by Eric Charles <er...@gmail.com>.
On 12/09/11 17:05, Jean-Daniel Cryans wrote:
> Also you referred to "George Lars", I don't understand those crazy
> Europeans that give their child two first names either (joking) but
> I'm 99.9999% sure it's the other way around :)
... and what to say about those Europeans who inherit a first name as
family name (smile)
> J-D
>
> On Mon, Sep 12, 2011 at 12:22 AM, Sujee Maniyam<su...@sujee.net> wrote:
>> Hi all
>> I have an article online, that shows how do some quick benchmarks on a HBase
>> cluster.
>> http://sujee.net/tech/articles/hbase-performance-testing/
>>
>> feedbacks / thoughts welcome.
>>
>> thanks very much
>> Sujee Maniyam
>> http://sujee.net
>>
Re: tutorial : HBase performance testing
Posted by Jean-Daniel Cryans <jd...@apache.org>.
Hi Sujee,
Both tools are lacking a step where they create the tables pre-split.
The difference can be staggering and possibly lead to misguiding.
Also you referred to "George Lars", I don't understand those crazy
Europeans that give their child two first names either (joking) but
I'm 99.9999% sure it's the other way around :)
J-D
On Mon, Sep 12, 2011 at 12:22 AM, Sujee Maniyam <su...@sujee.net> wrote:
> Hi all
> I have an article online, that shows how do some quick benchmarks on a HBase
> cluster.
> http://sujee.net/tech/articles/hbase-performance-testing/
>
> feedbacks / thoughts welcome.
>
> thanks very much
> Sujee Maniyam
> http://sujee.net
>