compare cassandra read n write results

[vineet daniel <vi...@gmail.com>]

Hi

A little while ago I tried cassandra's read n write operations and timed it.

I am using Pandra for communication with cassandra. System is CentOS 5 with
2 GB RAM and dual core.

I inserted 100000 rows in around 30 secs and read the same in 25 seconds.

If anyone of you have run similar tests can you please share or tell whether
this can be improved or not. I am using default configuration of cassandra
and its a single node setup.

Thanks
Vineet Daniel

Re: compare cassandra read n write results

[Jordan Pittier <jo...@gmail.com>]

First, read carefully and understand :
http://wiki.apache.org/cassandra/ThriftExamples#PHP

But you really shouldn't bother with benchmarks. Ask yourself this question
: "what if my Cassandra performs at 5k operation/s ? And what about 3k
op/s?". In other terms "why are you benchmarking ?". You've got 3000 op/s
with Pandra, last time I have checked I was arround 4k with a similar box.
All I know is it's enough for my needs and my future needs.

On Mon, Apr 12, 2010 at 4:41 PM, vineet daniel <vi...@gmail.com>wrote:

> Actually, to be honest I dont know how to insert 1000000 rows without PHP
> or Pandra. If you could help me out I will surely try it and will share the
> results with you guys.
>
> On Mon, Apr 12, 2010 at 7:25 PM, Paul Prescod <pr...@gmail.com> wrote:
>
>> How will they know whether the performance problem is caused by
>> Cassandra or Pandra if you do not have raw Cassandra performance
>> numbers for your setup?
>>
>> On Mon, Apr 12, 2010 at 5:51 AM, vineet daniel <vi...@gmail.com>
>> wrote:
>> > I dont think it would be a good idea not to use pandra for benchmarks as
>> we
>> > are going to use pandra for our application. Secondly, it will give
>> Pandra
>> > guys some boost to enhance the performance of thier library.
>>
>
>

Re: compare cassandra read n write results

[Paul Prescod <pa...@prescod.net>]

contrib/py_stress

Although that's still written in a scripting language, it at least
uses threading.

Anyhow, what's your real goal? Inserting 100K or 1M rows in 30 seconds
from a single-threaded environment like PHP is pretty good. Do your
business goals require more?

Also: Is it 100K or 1M? In one context you used 5 zeros and in another
you used 6.

On Mon, Apr 12, 2010 at 7:41 AM, vineet daniel <vi...@gmail.com> wrote:
> Actually, to be honest I dont know how to insert 1000000 rows without PHP or
> Pandra. If you could help me out I will surely try it and will share the
> results with you guys.

Re: compare cassandra read n write results

[vineet daniel <vi...@gmail.com>]

Actually, to be honest I dont know how to insert 1000000 rows without PHP or
Pandra. If you could help me out I will surely try it and will share the
results with you guys.

On Mon, Apr 12, 2010 at 7:25 PM, Paul Prescod <pr...@gmail.com> wrote:

> How will they know whether the performance problem is caused by
> Cassandra or Pandra if you do not have raw Cassandra performance
> numbers for your setup?
>
> On Mon, Apr 12, 2010 at 5:51 AM, vineet daniel <vi...@gmail.com>
> wrote:
> > I dont think it would be a good idea not to use pandra for benchmarks as
> we
> > are going to use pandra for our application. Secondly, it will give
> Pandra
> > guys some boost to enhance the performance of thier library.
>

Re: compare cassandra read n write results

[Paul Prescod <pr...@gmail.com>]

How will they know whether the performance problem is caused by
Cassandra or Pandra if you do not have raw Cassandra performance
numbers for your setup?

On Mon, Apr 12, 2010 at 5:51 AM, vineet daniel <vi...@gmail.com> wrote:
> I dont think it would be a good idea not to use pandra for benchmarks as we
> are going to use pandra for our application. Secondly, it will give Pandra
> guys some boost to enhance the performance of thier library.

Re: compare cassandra read n write results

[vineet daniel <vi...@gmail.com>]

I dont think it would be a good idea not to use pandra for benchmarks as we
are going to use pandra for our application. Secondly, it will give Pandra
guys some boost to enhance the performance of thier library.

On Mon, Apr 12, 2010 at 6:05 PM, Jordan Pittier <jo...@gmail.com>wrote:

> Hi,
> If you really want to benchmark your box, you should concidere not using
> Pandra nor any library built upon Thrift. They all come with a (small)
> overhead.
>
> I also realized when I made my first benchmark that most of my box's
> ressources was used by the benchmarking tool it self and not by Canssandra.
> I recommend using 2 boxes if possible, one for running the benchmark tool
> against the other which will run Cassandra (both boxes have to be in the
> same LAN).
>
> On Mon, Apr 12, 2010 at 1:55 PM, vineet daniel <vi...@gmail.com>wrote:
>
>> Hi
>>
>> A little while ago I tried cassandra's read n write operations and timed
>> it.
>> I am using Pandra for communication with cassandra. System is CentOS 5
>> with 2 GB RAM and dual core.
>>
>> I inserted 100000 rows in around 30 secs and read the same in 25 seconds.
>>
>> If anyone of you have run similar tests can you please share or tell
>> whether this can be improved or not. I am using default configuration of
>> cassandra and its a single node setup.
>>
>> Thanks
>> Vineet Daniel
>>
>
>

Re: compare cassandra read n write results

[Jordan Pittier <jo...@gmail.com>]

Hi,
If you really want to benchmark your box, you should concidere not using
Pandra nor any library built upon Thrift. They all come with a (small)
overhead.

I also realized when I made my first benchmark that most of my box's
ressources was used by the benchmarking tool it self and not by Canssandra.
I recommend using 2 boxes if possible, one for running the benchmark tool
against the other which will run Cassandra (both boxes have to be in the
same LAN).

On Mon, Apr 12, 2010 at 1:55 PM, vineet daniel <vi...@gmail.com>wrote:

> Hi
>
> A little while ago I tried cassandra's read n write operations and timed
> it.
> I am using Pandra for communication with cassandra. System is CentOS 5 with
> 2 GB RAM and dual core.
>
> I inserted 100000 rows in around 30 secs and read the same in 25 seconds.
>
> If anyone of you have run similar tests can you please share or tell
> whether this can be improved or not. I am using default configuration of
> cassandra and its a single node setup.
>
> Thanks
> Vineet Daniel
>

Re: compare cassandra read n write results

[Jonathan Ellis <jb...@gmail.com>]

If you use something standard like contrib/py_stress then maybe
comparing your numbers to ours would make sense.

On Mon, Apr 12, 2010 at 6:55 AM, vineet daniel <vi...@gmail.com> wrote:
> Hi
>
> A little while ago I tried cassandra's read n write operations and timed it.
> I am using Pandra for communication with cassandra. System is CentOS 5 with
> 2 GB RAM and dual core.
>
> I inserted 100000 rows in around 30 secs and read the same in 25 seconds.
>
> If anyone of you have run similar tests can you please share or tell whether
> this can be improved or not. I am using default configuration of cassandra
> and its a single node setup.
>
> Thanks
> Vineet Daniel
>