You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by SSam <to...@yahoo.com> on 2010/07/26 17:29:23 UTC
Cassandra Horizontal Scalability
>From Cassandra Website:
* Elastic
Read and write throughput both increase linearly as new machines are added,
with no downtime or interruption to applications.
I am testing TPS with Cassandra cluster.
Initially I have tested with one node cluster , got 1300 TPS,
added another node and I got 1200 TPS with 2 node cluster.
TPS supposed to increase with every additional node. I am not sure what I am
doing wrong? Any suggestions on Cassandra Scalability?
Thanks,
Sam.
Re: Cassandra Horizontal Scalability
Posted by SSam <to...@yahoo.com>.
Yes, I am sending requests to newly added node also.
Actually I am testing my web application with Cassandra, 1300 is application
TPS with Cassandra.
So far we are using MYSQL, planning to move use Cassandra, but I am getting
better TPS with MYSQL.
________________________________
From: Benjamin Black <b...@b3k.us>
To: user@cassandra.apache.org
Sent: Mon, July 26, 2010 12:50:07 PM
Subject: Re: Cassandra Horizontal Scalability
Are you sending all requests to a single server? Performance
increases when you add nodes when you send traffic to all of them at
once. Adding new ndoes does not cause the old nodes to speed up.
Also, 1300 reqs/sec is extremely slow, so something else is likely
wrong, as well.
On Mon, Jul 26, 2010 at 8:29 AM, SSam <to...@yahoo.com> wrote:
>
> From Cassandra Website:
>
> Elastic
>
> Read and write throughput both increase linearly as new machines are added,
> with no downtime or interruption to applications.
>
> I am testing TPS with Cassandra cluster.
> Initially I have tested with one node cluster , got 1300 TPS,
> added another node and I got 1200 TPS with 2 node cluster.
> TPS supposed to increase with every additional node. I am not sure what I
> am doing wrong? Any suggestions on Cassandra Scalability?
>
> Thanks,
> Sam.
>
>
>
Re: Cassandra Horizontal Scalability
Posted by Benjamin Black <b...@b3k.us>.
Are you sending all requests to a single server? Performance
increases when you add nodes when you send traffic to all of them at
once. Adding new ndoes does not cause the old nodes to speed up.
Also, 1300 reqs/sec is extremely slow, so something else is likely
wrong, as well.
On Mon, Jul 26, 2010 at 8:29 AM, SSam <to...@yahoo.com> wrote:
>
> From Cassandra Website:
>
> Elastic
>
> Read and write throughput both increase linearly as new machines are added,
> with no downtime or interruption to applications.
>
> I am testing TPS with Cassandra cluster.
> Initially I have tested with one node cluster , got 1300 TPS,
> added another node and I got 1200 TPS with 2 node cluster.
> TPS supposed to increase with every additional node. I am not sure what I
> am doing wrong? Any suggestions on Cassandra Scalability?
>
> Thanks,
> Sam.
>
>
>
Re: Cassandra Horizontal Scalability
Posted by Benjamin Black <b...@b3k.us>.
Are you persisting connections or opening a new one for each operation?
Do you understand what deletes actually do internally?
On Mon, Jul 26, 2010 at 12:45 PM, SSam <to...@yahoo.com> wrote:
> Thanks for reply
>
> I am doing very simple test:
> I am stressing my webapplication with grinder testing tool.
> Number of testing clients is 250, each client just creates/deletes a User
> Object( max 500 bytes data). So data is purely WRITE ( I hope Cassandra is
> Write Optimized) .
> I am running -0.6.3 version, no errors either on cassandra server logs or
> my web application server logs.
> I have JUST changed listing address and port from default storage-conf.xml
> Here is ColumnFamily Defination.
> <ColumnFamily Name="Users" CompareWith="UTF8Type"/>
>
> Anyway I will do testing directly over cassandra bypassing web application.
> I will update the forum with my testing results.
>
>
> ________________________________
> From: Bill de hÓra <bi...@dehora.net>
> To: user@cassandra.apache.org
> Sent: Mon, July 26, 2010 3:23:38 PM
> Subject: Re: Cassandra Horizontal Scalability
>
> Initially, try running stress.py against your servers. That will help
> eliminate pelops as the issue.
>
> Also, some more detail on your setup wouldn't hurt, 1300 anything a
> second is low.
>
> Is the TPS reads or writes, or a mix?
>
> What's the column family structure?
>
> Is the data large (eg are you writing binary files)?
>
> How many client processes are you running?
>
> What are the server configurations?
>
> How much data are nodes carrying?
>
> What does the output of ./nodetool cfstats say?
>
> What version of cassandra are you running?
>
> Are there errors in the server logs?
>
> Bill
>
> On Mon, 2010-07-26 at 09:26 -0700, SSam wrote:
>>
>> I have fully separated gigabit private subnet.
>> I am using pelops cassandra java client with 150 client pool.
>
>>
>>
>>
>> ______________________________________________________________________
>> From: Paul Prescod <pa...@prescod.net>
>> To: user@cassandra.apache.org
>> Sent: Mon, July 26, 2010 12:21:06 PM
>> Subject: Re: Cassandra Horizontal Scalability
>>
>> There are a lot of variables that go into a proper benchmark. The
>> bottleneck could be in many different places.
>>
>>
>> How many client threads are you using? What kind of network?
>>
>> On Mon, Jul 26, 2010 at 8:29 AM, SSam <to...@yahoo.com> wrote:
>>
>> From Cassandra Website:
>> * Elastic
>>
>> Read and write throughput both increase linearly as
>> new machines are added, with no downtime or
>> interruption to applications.
>>
>>
>> I am testing TPS with Cassandra cluster.
>> Initially I have tested with one node cluster , got 1300
>> TPS,
>> added another node and I got 1200 TPS with 2 node cluster.
>> TPS supposed to increase with every additional node. I am
>> not sure what I am doing wrong? Any suggestions on Cassandra
>> Scalability?
>>
>> Thanks,
>> Sam.
>>
>>
>>
>>
>>
>>
>>
>>
>
>
>
>
Re: Cassandra Horizontal Scalability
Posted by SSam <to...@yahoo.com>.
Thanks for reply
I am doing very simple test:
I am stressing my webapplication with grinder testing tool.
Number of testing clients is 250, each client just creates/deletes a User
Object( max 500 bytes data). So data is purely WRITE ( I hope Cassandra is
Write Optimized) .
I am running -0.6.3 version, no errors either on cassandra server logs or my
web application server logs.
I have JUST changed listing address and port from default storage-conf.xml
Here is ColumnFamily Defination.
<ColumnFamily Name="Users" CompareWith="UTF8Type"/>
Anyway I will do testing directly over cassandra bypassing web application. I
will update the forum with my testing results.
________________________________
From: Bill de hÓra <bi...@dehora.net>
To: user@cassandra.apache.org
Sent: Mon, July 26, 2010 3:23:38 PM
Subject: Re: Cassandra Horizontal Scalability
Initially, try running stress.py against your servers. That will help
eliminate pelops as the issue.
Also, some more detail on your setup wouldn't hurt, 1300 anything a
second is low.
Is the TPS reads or writes, or a mix?
What's the column family structure?
Is the data large (eg are you writing binary files)?
How many client processes are you running?
What are the server configurations?
How much data are nodes carrying?
What does the output of ./nodetool cfstats say?
What version of cassandra are you running?
Are there errors in the server logs?
Bill
On Mon, 2010-07-26 at 09:26 -0700, SSam wrote:
>
> I have fully separated gigabit private subnet.
> I am using pelops cassandra java client with 150 client pool.
>
>
>
> ______________________________________________________________________
> From: Paul Prescod <pa...@prescod.net>
> To: user@cassandra.apache.org
> Sent: Mon, July 26, 2010 12:21:06 PM
> Subject: Re: Cassandra Horizontal Scalability
>
> There are a lot of variables that go into a proper benchmark. The
> bottleneck could be in many different places.
>
>
> How many client threads are you using? What kind of network?
>
> On Mon, Jul 26, 2010 at 8:29 AM, SSam <to...@yahoo.com> wrote:
>
> From Cassandra Website:
> * Elastic
>
> Read and write throughput both increase linearly as
> new machines are added, with no downtime or
> interruption to applications.
>
>
> I am testing TPS with Cassandra cluster.
> Initially I have tested with one node cluster , got 1300
> TPS,
> added another node and I got 1200 TPS with 2 node cluster.
> TPS supposed to increase with every additional node. I am
> not sure what I am doing wrong? Any suggestions on Cassandra
> Scalability?
>
> Thanks,
> Sam.
>
>
>
>
>
>
>
>
Re: Cassandra Horizontal Scalability
Posted by Bill de hÓra <bi...@dehora.net>.
Initially, try running stress.py against your servers. That will help
eliminate pelops as the issue.
Also, some more detail on your setup wouldn't hurt, 1300 anything a
second is low.
Is the TPS reads or writes, or a mix?
What's the column family structure?
Is the data large (eg are you writing binary files)?
How many client processes are you running?
What are the server configurations?
How much data are nodes carrying?
What does the output of ./nodetool cfstats say?
What version of cassandra are you running?
Are there errors in the server logs?
Bill
On Mon, 2010-07-26 at 09:26 -0700, SSam wrote:
>
> I have fully separated gigabit private subnet.
> I am using pelops cassandra java client with 150 client pool.
>
>
>
> ______________________________________________________________________
> From: Paul Prescod <pa...@prescod.net>
> To: user@cassandra.apache.org
> Sent: Mon, July 26, 2010 12:21:06 PM
> Subject: Re: Cassandra Horizontal Scalability
>
> There are a lot of variables that go into a proper benchmark. The
> bottleneck could be in many different places.
>
>
> How many client threads are you using? What kind of network?
>
> On Mon, Jul 26, 2010 at 8:29 AM, SSam <to...@yahoo.com> wrote:
>
> From Cassandra Website:
> * Elastic
>
> Read and write throughput both increase linearly as
> new machines are added, with no downtime or
> interruption to applications.
>
>
> I am testing TPS with Cassandra cluster.
> Initially I have tested with one node cluster , got 1300
> TPS,
> added another node and I got 1200 TPS with 2 node cluster.
> TPS supposed to increase with every additional node. I am
> not sure what I am doing wrong? Any suggestions on Cassandra
> Scalability?
>
> Thanks,
> Sam.
>
>
>
>
>
>
>
>
Re: Cassandra Horizontal Scalability
Posted by SSam <to...@yahoo.com>.
I have fully separated gigabit private subnet.
I am using pelops cassandra java client with 150 client pool.
________________________________
From: Paul Prescod <pa...@prescod.net>
To: user@cassandra.apache.org
Sent: Mon, July 26, 2010 12:21:06 PM
Subject: Re: Cassandra Horizontal Scalability
There are a lot of variables that go into a proper benchmark. The bottleneck
could be in many different places.
How many client threads are you using? What kind of network?
On Mon, Jul 26, 2010 at 8:29 AM, SSam <to...@yahoo.com> wrote:
>
>>From Cassandra Website:
>
> * Elastic
>Read and write throughput both increase linearly as new machines are added,
>with no downtime or interruption to applications.
>
>I am testing TPS with Cassandra cluster.
>Initially I have tested with one node cluster , got 1300 TPS,
>added another node and I got 1200 TPS with 2 node cluster.
>TPS supposed to increase with every additional node. I am not sure what I am
>doing wrong? Any suggestions on Cassandra Scalability?
>
>
>Thanks,
>Sam.
>
>
>
Re: Cassandra Horizontal Scalability
Posted by Paul Prescod <pa...@prescod.net>.
There are a lot of variables that go into a proper benchmark. The bottleneck
could be in many different places.
How many client threads are you using? What kind of network?
On Mon, Jul 26, 2010 at 8:29 AM, SSam <to...@yahoo.com> wrote:
>
> From Cassandra Website:
>
> - *Elastic*
>
> Read and write throughput both increase linearly as new machines are
> added, with no downtime or interruption to applications.
>
>
> I am testing TPS with Cassandra cluster.
> Initially I have tested with one node cluster , got 1300 TPS,
> added another node and I got 1200 TPS with 2 node cluster.
> TPS supposed to increase with every additional node. I am not sure what I
> am doing wrong? Any suggestions on Cassandra Scalability?
>
> Thanks,
> Sam.
>
>
>