You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@gora.apache.org by Sheriffo Ceesay <sn...@gmail.com> on 2019/07/21 16:32:40 UTC

Week 8 Report

Week eight report is available at
https://cwiki.apache.org/confluence/display/GORA/%5BGORA-532%5D+Apache+Gora+Benchmark+Module+Weekly+Report

I ran some workloads to compare Gora implementation of Mongo and HBase to
the native implementations. Plots are provided in the reports and the
generated data is also available at [1]. This is a work in progress, I have
done Workload A and Workload B [2].

Please let me know if you have any suggestions or questions.

[1] https://github.com/sneceesay77/gora/tree/GORA-532
[2] https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads


**Sheriffo Ceesay**

Re: Week 8 Report

Posted by Sheriffo Ceesay <sn...@gmail.com>.
**Sheriffo Ceesay**
Hi Furkan,

Thanks for the reply. I have replied to your comment inline.

On Tue, Jul 23, 2019 at 1:10 AM Furkan KAMACI <fu...@gmail.com>
wrote:

> Hi Sheriffo,
>
> I've checked all the GSoC reports including your's. Thanks for filling your
> reports with time slot information about tasks.
>
> You have 3 benchmark result charts at your Week 8 Report. Hbase-native is
> dramatically slow compared to the others at 2 out of 3. Do you have any
> comment about it?


Thanks for the feedback. Right now I don't have a concrete explanation but
I suspect HBase setup. I am using the default setup for all datastores
without setting any optimization settings in the configuration files. I
will have a look at this week. My only concern is I don't want to over
optimise the configuration of a particular datastore thereby giving an
unfair advantage over the others.

Thank you.

>
> Kind Regards,
> Furkan KAMACI
>
> On Sun, Jul 21, 2019 at 7:33 PM Sheriffo Ceesay <sn...@gmail.com>
> wrote:
>
> > Week eight report is available at
> >
> >
> https://cwiki.apache.org/confluence/display/GORA/%5BGORA-532%5D+Apache+Gora+Benchmark+Module+Weekly+Report
> >
> > I ran some workloads to compare Gora implementation of Mongo and HBase to
> > the native implementations. Plots are provided in the reports and the
> > generated data is also available at [1]. This is a work in progress, I
> have
> > done Workload A and Workload B [2].
> >
> > Please let me know if you have any suggestions or questions.
> >
> > [1] https://github.com/sneceesay77/gora/tree/GORA-532
> > [2] https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
> >
> >
> > **Sheriffo Ceesay**
> >
>

Re: Week 8 Report

Posted by Sheriffo Ceesay <sn...@gmail.com>.
Hi Kevin,

Thanks again for the input.

I have taken note of your comments and I will factor them in
moving forward.

Thank you.


**Sheriffo Ceesay**


On Tue, Jul 23, 2019 at 10:13 AM Kevin Ratnasekera <dj...@gmail.com>
wrote:

> Hi Sheriffo,
>
> Thank you for your findings and hard work on this. I agree on most of the
> points you already mentioned. But I dont think we have consistent client
> implementations across all data stores in Apache Gora. That means some
> clients use async version of their API s as default, some uses sync. Some
> clients have connection pooling implemented and some use single connection
> to do all the data store work. These configurations will generally change
> the behavior when these client s are performed under a huge load. I think
> we will be good, if we do track the client setup/configurations in which
> benchmarks are captured.
>
> Regards
> Kevin
>
> On Tue, Jul 23, 2019 at 2:14 PM Sheriffo Ceesay <sn...@gmail.com>
> wrote:
>
> > **Sheriffo Ceesay**
> >
> >
> > On Tue, Jul 23, 2019 at 6:52 AM Kevin Ratnasekera <
> djkevincr1989@gmail.com
> > >
> > wrote:
> >
> > > Hi Sheriffo,
> > >
> > > Adding to what Kamaci already mentioned, Have you tried Hbase-Store
> with
> > > buffered mutator engaged? [1] It allows HBase operations Eg:- puts to
> be
> > > batched and asynchronous.
> >
> >
> > Hi Kevin,
> > Yes I am using *gora.hbasestore.hbase.client.autoflush.enabled=true *in
> the
> > gora.properties file.
> >
> >
> >
> > > On related note also have a look on points where
> > > you flush() the datastore with native HBase implementation.
> > >
> >
> > The native implementation is through Yahoo! Cloud Service Benchmark. I
> > haven't gone through their implementation. I think it may be due to the
> > default configuration that I am using. I will try to dig further to see
> if
> > I can strike some improvement.
> >
> >
> > >
> > > [1] gora.hbasestore.hbase.client.autoflush.enabled=true
> > >
> > > Regards
> > > Kevin
> > >
> > > On Tue, Jul 23, 2019 at 5:40 AM Furkan KAMACI <fu...@gmail.com>
> > > wrote:
> > >
> > > > Hi Sheriffo,
> > > >
> > > > I've checked all the GSoC reports including your's. Thanks for
> filling
> > > > your reports with time slot information about tasks.
> > > >
> > > > You have 3 benchmark result charts at your Week 8 Report.
> Hbase-native
> > is
> > > > dramatically slow compared to the others at 2 out of 3. Do you have
> any
> > > > comment about it?
> > > >
> > > > Kind Regards,
> > > > Furkan KAMACI
> > > >
> > > > On Sun, Jul 21, 2019 at 7:33 PM Sheriffo Ceesay <
> sneceesay77@gmail.com
> > >
> > > > wrote:
> > > >
> > > >> Week eight report is available at
> > > >>
> > > >>
> > >
> >
> https://cwiki.apache.org/confluence/display/GORA/%5BGORA-532%5D+Apache+Gora+Benchmark+Module+Weekly+Report
> > > >>
> > > >> I ran some workloads to compare Gora implementation of Mongo and
> HBase
> > > to
> > > >> the native implementations. Plots are provided in the reports and
> the
> > > >> generated data is also available at [1]. This is a work in
> progress, I
> > > >> have
> > > >> done Workload A and Workload B [2].
> > > >>
> > > >> Please let me know if you have any suggestions or questions.
> > > >>
> > > >> [1] https://github.com/sneceesay77/gora/tree/GORA-532
> > > >> [2] https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
> > > >>
> > > >>
> > > >> **Sheriffo Ceesay**
> > > >>
> > > >
> > >
> >
>

Re: Week 8 Report

Posted by Kevin Ratnasekera <dj...@gmail.com>.
On related note: Please also make on note on the client version of the
libraries we are testing for both native and gora implementations.

On Tue, Jul 23, 2019 at 2:43 PM Kevin Ratnasekera <dj...@gmail.com>
wrote:

> Hi Sheriffo,
>
> Thank you for your findings and hard work on this. I agree on most of the
> points you already mentioned. But I dont think we have consistent client
> implementations across all data stores in Apache Gora. That means some
> clients use async version of their API s as default, some uses sync. Some
> clients have connection pooling implemented and some use single connection
> to do all the data store work. These configurations will generally change
> the behavior when these client s are performed under a huge load. I think
> we will be good, if we do track the client setup/configurations in which
> benchmarks are captured.
>
> Regards
> Kevin
>
> On Tue, Jul 23, 2019 at 2:14 PM Sheriffo Ceesay <sn...@gmail.com>
> wrote:
>
>> **Sheriffo Ceesay**
>>
>>
>> On Tue, Jul 23, 2019 at 6:52 AM Kevin Ratnasekera <
>> djkevincr1989@gmail.com>
>> wrote:
>>
>> > Hi Sheriffo,
>> >
>> > Adding to what Kamaci already mentioned, Have you tried Hbase-Store with
>> > buffered mutator engaged? [1] It allows HBase operations Eg:- puts to be
>> > batched and asynchronous.
>>
>>
>> Hi Kevin,
>> Yes I am using *gora.hbasestore.hbase.client.autoflush.enabled=true *in
>> the
>> gora.properties file.
>>
>>
>>
>> > On related note also have a look on points where
>> > you flush() the datastore with native HBase implementation.
>> >
>>
>> The native implementation is through Yahoo! Cloud Service Benchmark. I
>> haven't gone through their implementation. I think it may be due to the
>> default configuration that I am using. I will try to dig further to see if
>> I can strike some improvement.
>>
>>
>> >
>> > [1] gora.hbasestore.hbase.client.autoflush.enabled=true
>> >
>> > Regards
>> > Kevin
>> >
>> > On Tue, Jul 23, 2019 at 5:40 AM Furkan KAMACI <fu...@gmail.com>
>> > wrote:
>> >
>> > > Hi Sheriffo,
>> > >
>> > > I've checked all the GSoC reports including your's. Thanks for filling
>> > > your reports with time slot information about tasks.
>> > >
>> > > You have 3 benchmark result charts at your Week 8 Report.
>> Hbase-native is
>> > > dramatically slow compared to the others at 2 out of 3. Do you have
>> any
>> > > comment about it?
>> > >
>> > > Kind Regards,
>> > > Furkan KAMACI
>> > >
>> > > On Sun, Jul 21, 2019 at 7:33 PM Sheriffo Ceesay <
>> sneceesay77@gmail.com>
>> > > wrote:
>> > >
>> > >> Week eight report is available at
>> > >>
>> > >>
>> >
>> https://cwiki.apache.org/confluence/display/GORA/%5BGORA-532%5D+Apache+Gora+Benchmark+Module+Weekly+Report
>> > >>
>> > >> I ran some workloads to compare Gora implementation of Mongo and
>> HBase
>> > to
>> > >> the native implementations. Plots are provided in the reports and the
>> > >> generated data is also available at [1]. This is a work in progress,
>> I
>> > >> have
>> > >> done Workload A and Workload B [2].
>> > >>
>> > >> Please let me know if you have any suggestions or questions.
>> > >>
>> > >> [1] https://github.com/sneceesay77/gora/tree/GORA-532
>> > >> [2] https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
>> > >>
>> > >>
>> > >> **Sheriffo Ceesay**
>> > >>
>> > >
>> >
>>
>

Re: Week 8 Report

Posted by Kevin Ratnasekera <dj...@gmail.com>.
Hi Sheriffo,

Thank you for your findings and hard work on this. I agree on most of the
points you already mentioned. But I dont think we have consistent client
implementations across all data stores in Apache Gora. That means some
clients use async version of their API s as default, some uses sync. Some
clients have connection pooling implemented and some use single connection
to do all the data store work. These configurations will generally change
the behavior when these client s are performed under a huge load. I think
we will be good, if we do track the client setup/configurations in which
benchmarks are captured.

Regards
Kevin

On Tue, Jul 23, 2019 at 2:14 PM Sheriffo Ceesay <sn...@gmail.com>
wrote:

> **Sheriffo Ceesay**
>
>
> On Tue, Jul 23, 2019 at 6:52 AM Kevin Ratnasekera <djkevincr1989@gmail.com
> >
> wrote:
>
> > Hi Sheriffo,
> >
> > Adding to what Kamaci already mentioned, Have you tried Hbase-Store with
> > buffered mutator engaged? [1] It allows HBase operations Eg:- puts to be
> > batched and asynchronous.
>
>
> Hi Kevin,
> Yes I am using *gora.hbasestore.hbase.client.autoflush.enabled=true *in the
> gora.properties file.
>
>
>
> > On related note also have a look on points where
> > you flush() the datastore with native HBase implementation.
> >
>
> The native implementation is through Yahoo! Cloud Service Benchmark. I
> haven't gone through their implementation. I think it may be due to the
> default configuration that I am using. I will try to dig further to see if
> I can strike some improvement.
>
>
> >
> > [1] gora.hbasestore.hbase.client.autoflush.enabled=true
> >
> > Regards
> > Kevin
> >
> > On Tue, Jul 23, 2019 at 5:40 AM Furkan KAMACI <fu...@gmail.com>
> > wrote:
> >
> > > Hi Sheriffo,
> > >
> > > I've checked all the GSoC reports including your's. Thanks for filling
> > > your reports with time slot information about tasks.
> > >
> > > You have 3 benchmark result charts at your Week 8 Report. Hbase-native
> is
> > > dramatically slow compared to the others at 2 out of 3. Do you have any
> > > comment about it?
> > >
> > > Kind Regards,
> > > Furkan KAMACI
> > >
> > > On Sun, Jul 21, 2019 at 7:33 PM Sheriffo Ceesay <sneceesay77@gmail.com
> >
> > > wrote:
> > >
> > >> Week eight report is available at
> > >>
> > >>
> >
> https://cwiki.apache.org/confluence/display/GORA/%5BGORA-532%5D+Apache+Gora+Benchmark+Module+Weekly+Report
> > >>
> > >> I ran some workloads to compare Gora implementation of Mongo and HBase
> > to
> > >> the native implementations. Plots are provided in the reports and the
> > >> generated data is also available at [1]. This is a work in progress, I
> > >> have
> > >> done Workload A and Workload B [2].
> > >>
> > >> Please let me know if you have any suggestions or questions.
> > >>
> > >> [1] https://github.com/sneceesay77/gora/tree/GORA-532
> > >> [2] https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
> > >>
> > >>
> > >> **Sheriffo Ceesay**
> > >>
> > >
> >
>

Re: Week 8 Report

Posted by Sheriffo Ceesay <sn...@gmail.com>.
**Sheriffo Ceesay**


On Tue, Jul 23, 2019 at 6:52 AM Kevin Ratnasekera <dj...@gmail.com>
wrote:

> Hi Sheriffo,
>
> Adding to what Kamaci already mentioned, Have you tried Hbase-Store with
> buffered mutator engaged? [1] It allows HBase operations Eg:- puts to be
> batched and asynchronous.


Hi Kevin,
Yes I am using *gora.hbasestore.hbase.client.autoflush.enabled=true *in the
gora.properties file.



> On related note also have a look on points where
> you flush() the datastore with native HBase implementation.
>

The native implementation is through Yahoo! Cloud Service Benchmark. I
haven't gone through their implementation. I think it may be due to the
default configuration that I am using. I will try to dig further to see if
I can strike some improvement.


>
> [1] gora.hbasestore.hbase.client.autoflush.enabled=true
>
> Regards
> Kevin
>
> On Tue, Jul 23, 2019 at 5:40 AM Furkan KAMACI <fu...@gmail.com>
> wrote:
>
> > Hi Sheriffo,
> >
> > I've checked all the GSoC reports including your's. Thanks for filling
> > your reports with time slot information about tasks.
> >
> > You have 3 benchmark result charts at your Week 8 Report. Hbase-native is
> > dramatically slow compared to the others at 2 out of 3. Do you have any
> > comment about it?
> >
> > Kind Regards,
> > Furkan KAMACI
> >
> > On Sun, Jul 21, 2019 at 7:33 PM Sheriffo Ceesay <sn...@gmail.com>
> > wrote:
> >
> >> Week eight report is available at
> >>
> >>
> https://cwiki.apache.org/confluence/display/GORA/%5BGORA-532%5D+Apache+Gora+Benchmark+Module+Weekly+Report
> >>
> >> I ran some workloads to compare Gora implementation of Mongo and HBase
> to
> >> the native implementations. Plots are provided in the reports and the
> >> generated data is also available at [1]. This is a work in progress, I
> >> have
> >> done Workload A and Workload B [2].
> >>
> >> Please let me know if you have any suggestions or questions.
> >>
> >> [1] https://github.com/sneceesay77/gora/tree/GORA-532
> >> [2] https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
> >>
> >>
> >> **Sheriffo Ceesay**
> >>
> >
>

Re: Week 8 Report

Posted by Kevin Ratnasekera <dj...@gmail.com>.
Hi Sheriffo,

Adding to what Kamaci already mentioned, Have you tried Hbase-Store with
buffered mutator engaged? [1] It allows HBase operations Eg:- puts to be
batched and asynchronous. On related note also have a look on points where
you flush() the datastore with native HBase implementation.

[1] gora.hbasestore.hbase.client.autoflush.enabled=true

Regards
Kevin

On Tue, Jul 23, 2019 at 5:40 AM Furkan KAMACI <fu...@gmail.com>
wrote:

> Hi Sheriffo,
>
> I've checked all the GSoC reports including your's. Thanks for filling
> your reports with time slot information about tasks.
>
> You have 3 benchmark result charts at your Week 8 Report. Hbase-native is
> dramatically slow compared to the others at 2 out of 3. Do you have any
> comment about it?
>
> Kind Regards,
> Furkan KAMACI
>
> On Sun, Jul 21, 2019 at 7:33 PM Sheriffo Ceesay <sn...@gmail.com>
> wrote:
>
>> Week eight report is available at
>>
>> https://cwiki.apache.org/confluence/display/GORA/%5BGORA-532%5D+Apache+Gora+Benchmark+Module+Weekly+Report
>>
>> I ran some workloads to compare Gora implementation of Mongo and HBase to
>> the native implementations. Plots are provided in the reports and the
>> generated data is also available at [1]. This is a work in progress, I
>> have
>> done Workload A and Workload B [2].
>>
>> Please let me know if you have any suggestions or questions.
>>
>> [1] https://github.com/sneceesay77/gora/tree/GORA-532
>> [2] https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
>>
>>
>> **Sheriffo Ceesay**
>>
>

Re: Week 8 Report

Posted by Furkan KAMACI <fu...@gmail.com>.
Hi Sheriffo,

I've checked all the GSoC reports including your's. Thanks for filling your
reports with time slot information about tasks.

You have 3 benchmark result charts at your Week 8 Report. Hbase-native is
dramatically slow compared to the others at 2 out of 3. Do you have any
comment about it?

Kind Regards,
Furkan KAMACI

On Sun, Jul 21, 2019 at 7:33 PM Sheriffo Ceesay <sn...@gmail.com>
wrote:

> Week eight report is available at
>
> https://cwiki.apache.org/confluence/display/GORA/%5BGORA-532%5D+Apache+Gora+Benchmark+Module+Weekly+Report
>
> I ran some workloads to compare Gora implementation of Mongo and HBase to
> the native implementations. Plots are provided in the reports and the
> generated data is also available at [1]. This is a work in progress, I have
> done Workload A and Workload B [2].
>
> Please let me know if you have any suggestions or questions.
>
> [1] https://github.com/sneceesay77/gora/tree/GORA-532
> [2] https://github.com/brianfrankcooper/YCSB/wiki/Core-Workloads
>
>
> **Sheriffo Ceesay**
>