You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@activemq.apache.org by kal123 <kp...@gmail.com> on 2014/02/11 17:32:32 UTC
performance Replicated LevelDB vs. SAN
The following links states:
http://activemq.apache.org/masterslave.html
For those willing to try out new tech, the Replicated LevelDB Store gives
speeds similar to a SAN solution without the hassle of having to setup a
highly available shared file system.
During our testing with Replicated LevelDB, the hugh difference between the
SAN and Replicated Leveldb
With one queue we observed ~6000 msg/per sec with SAN, while replicated
leveldb was about ~600 msg/per sec.
Does anyone have any input on performance and if any tuning had to be done
to achieve it?
--
View this message in context: http://activemq.2283324.n4.nabble.com/performance-Replicated-LevelDB-vs-SAN-tp4677811.html
Sent from the ActiveMQ - User mailing list archive at Nabble.com.
Re: performance Replicated LevelDB vs. SAN
Posted by Chirag Pujara <ch...@gmail.com>.
Hello kal,
I am also trying to setup activemq with LevelDb. But having issues and
Activemq stops processing messages on queue. I can't figure out issue. Can
you share your settings/configuration for zookeeper and activemq so that I
can figure out what is going on.
thanks,
chirag
On Tue, Feb 11, 2014 at 10:32 AM, kal123 <kp...@gmail.com> wrote:
> The following links states:
> http://activemq.apache.org/masterslave.html
> For those willing to try out new tech, the Replicated LevelDB Store gives
> speeds similar to a SAN solution without the hassle of having to setup a
> highly available shared file system.
>
>
> During our testing with Replicated LevelDB, the hugh difference between the
> SAN and Replicated Leveldb
>
> With one queue we observed ~6000 msg/per sec with SAN, while replicated
> leveldb was about ~600 msg/per sec.
>
> Does anyone have any input on performance and if any tuning had to be done
> to achieve it?
>
>
>
>
> --
> View this message in context:
> http://activemq.2283324.n4.nabble.com/performance-Replicated-LevelDB-vs-SAN-tp4677811.html
> Sent from the ActiveMQ - User mailing list archive at Nabble.com.
>
Re: performance Replicated LevelDB vs. SAN
Posted by Noel OConnor <no...@gmail.com>.
Check if you're hitting the max disk write on your local disk.
SAN disks can have battery backed caches which write to memory before
writing to disk.
On Wed, Feb 12, 2014 at 3:32 AM, kal123 <kp...@gmail.com> wrote:
> The following links states:
> http://activemq.apache.org/masterslave.html
> For those willing to try out new tech, the Replicated LevelDB Store gives
> speeds similar to a SAN solution without the hassle of having to setup a
> highly available shared file system.
>
>
> During our testing with Replicated LevelDB, the hugh difference between the
> SAN and Replicated Leveldb
>
> With one queue we observed ~6000 msg/per sec with SAN, while replicated
> leveldb was about ~600 msg/per sec.
>
> Does anyone have any input on performance and if any tuning had to be done
> to achieve it?
>
>
>
>
> --
> View this message in context:
> http://activemq.2283324.n4.nabble.com/performance-Replicated-LevelDB-vs-SAN-tp4677811.html
> Sent from the ActiveMQ - User mailing list archive at Nabble.com.
>
Re: performance Replicated LevelDB vs. SAN
Posted by Oleg Dulin <ol...@gmail.com>.
I just wanted to add that when I am not using replication it works swimmingly.
But I'd really like to get replication working today. It is the main
reason for the upgrade for me.
On 2014-03-25 17:15:18 +0000, Claus Ibsen said:
> There has been numerous fixes and improves for leveldb on the 5.10
> branch. So you may want to try building from latest source code and
> try with a SNAPSHOT of 5.10.
>
> On Tue, Mar 25, 2014 at 5:55 PM, Oleg Dulin
> <ol...@gmail.com> wrote:
>> I am running a similar test.
>>
>> Replicated LevelDB, 3 Zookepers, 3 AMQ brokers with local_mem sync, I am
>> publishing messages on a queue using one thread, and taking them off that
>> queue on another thread.
>>
>> Performance is abysmal, 1700 messages or so go out pretty quick, but then it
>> pauses every 50 messages or so.
>>
>> The settings are mostly taken from the sample Replicated LevelDB
>> configuration.
>>
>> What gives ?
>>
>> Any input is greatly appreciated.
>>
>>
>> On 2014-02-14 20:26:41 +0000, shippers said:
>>
>>> I'm testing with ActiveMQ Replicated too (latest SNAPSHOT build) and
>>> seeing
>>> similar problems with stability and reliability. Performance-wise, it
>>> depends on the messaging pattern, number of pubs / subs, message size, and
>>> transaction size. Also, very dependent on network speed and local disk
>>> speed.
>>>
>>> Can you describe your testing env? Size of messages? # of clients pub
>>> and
>>> sub? Network speed?
>>>
>>> I'm running a 3 node ActiveMQ quorum_mem and 3 node Zookeeper.
>>>
>>> Example test, 1 pub and 1 sub client running on different hardware, 4k
>>> message size, persisted messages, not using transactions, 10G network,
>>> virtual hardware box of 2 CPUs and 8 GIGs memory.
>>>
>>> ~900 msgs/per sec steady state (pub / sub running normally, keeping up
>>> with
>>> each other)
>>
>>
>>
>> --
>> Regards,
>> Oleg Dulin
>> http://www.olegdulin.com
--
Regards,
Oleg Dulin
http://www.olegdulin.com
Re: performance Replicated LevelDB vs. SAN
Posted by Oleg Dulin <ol...@gmail.com>.
Claus:
Unfortunately I don't have the energy and time to mess with snapshot
builds... I'd love to though.
The specific problem I have is that after about 1700 messages the whole
thing slows down to a crawl. It doesn't seem to be the case if I am
using plain leveldb, but when using replication I come across problems.
Is there some configuration I am missing ? This is all there is to it:
94 <persistenceAdapter>
95 <replicatedLevelDB
96 directory="${activemq.data}"
97 replicas="3"
98 bind="tcp://192.168.7.107:3201"
99 zkAddress="127.0.0.1:2181"
100 zkPassword="password"
101 zkPath="/activemq/leveldb-stores"
102 sync="quorum_mem"
103 />
104 </persistenceAdapter>
On 2014-03-25 17:15:18 +0000, Claus Ibsen said:
> There has been numerous fixes and improves for leveldb on the 5.10
> branch. So you may want to try building from latest source code and
> try with a SNAPSHOT of 5.10.
>
> On Tue, Mar 25, 2014 at 5:55 PM, Oleg Dulin
> <ol...@gmail.com> wrote:
>> I am running a similar test.
>>
>> Replicated LevelDB, 3 Zookepers, 3 AMQ brokers with local_mem sync, I am
>> publishing messages on a queue using one thread, and taking them off that
>> queue on another thread.
>>
>> Performance is abysmal, 1700 messages or so go out pretty quick, but then it
>> pauses every 50 messages or so.
>>
>> The settings are mostly taken from the sample Replicated LevelDB
>> configuration.
>>
>> What gives ?
>>
>> Any input is greatly appreciated.
>>
>>
>> On 2014-02-14 20:26:41 +0000, shippers said:
>>
>>> I'm testing with ActiveMQ Replicated too (latest SNAPSHOT build) and
>>> seeing
>>> similar problems with stability and reliability. Performance-wise, it
>>> depends on the messaging pattern, number of pubs / subs, message size, and
>>> transaction size. Also, very dependent on network speed and local disk
>>> speed.
>>>
>>> Can you describe your testing env? Size of messages? # of clients pub
>>> and
>>> sub? Network speed?
>>>
>>> I'm running a 3 node ActiveMQ quorum_mem and 3 node Zookeeper.
>>>
>>> Example test, 1 pub and 1 sub client running on different hardware, 4k
>>> message size, persisted messages, not using transactions, 10G network,
>>> virtual hardware box of 2 CPUs and 8 GIGs memory.
>>>
>>> ~900 msgs/per sec steady state (pub / sub running normally, keeping up
>>> with
>>> each other)
>>
>>
>>
>> --
>> Regards,
>> Oleg Dulin
>> http://www.olegdulin.com
--
Regards,
Oleg Dulin
http://www.olegdulin.com
Re: performance Replicated LevelDB vs. SAN
Posted by Claus Ibsen <cl...@gmail.com>.
There has been numerous fixes and improves for leveldb on the 5.10
branch. So you may want to try building from latest source code and
try with a SNAPSHOT of 5.10.
On Tue, Mar 25, 2014 at 5:55 PM, Oleg Dulin <ol...@gmail.com> wrote:
> I am running a similar test.
>
> Replicated LevelDB, 3 Zookepers, 3 AMQ brokers with local_mem sync, I am
> publishing messages on a queue using one thread, and taking them off that
> queue on another thread.
>
> Performance is abysmal, 1700 messages or so go out pretty quick, but then it
> pauses every 50 messages or so.
>
> The settings are mostly taken from the sample Replicated LevelDB
> configuration.
>
> What gives ?
>
> Any input is greatly appreciated.
>
>
> On 2014-02-14 20:26:41 +0000, shippers said:
>
>> I'm testing with ActiveMQ Replicated too (latest SNAPSHOT build) and
>> seeing
>> similar problems with stability and reliability. Performance-wise, it
>> depends on the messaging pattern, number of pubs / subs, message size, and
>> transaction size. Also, very dependent on network speed and local disk
>> speed.
>>
>> Can you describe your testing env? Size of messages? # of clients pub
>> and
>> sub? Network speed?
>>
>> I'm running a 3 node ActiveMQ quorum_mem and 3 node Zookeeper.
>>
>> Example test, 1 pub and 1 sub client running on different hardware, 4k
>> message size, persisted messages, not using transactions, 10G network,
>> virtual hardware box of 2 CPUs and 8 GIGs memory.
>>
>> ~900 msgs/per sec steady state (pub / sub running normally, keeping up
>> with
>> each other)
>
>
>
> --
> Regards,
> Oleg Dulin
> http://www.olegdulin.com
>
>
--
Claus Ibsen
-----------------
Red Hat, Inc.
Email: cibsen@redhat.com
Twitter: davsclaus
Blog: http://davsclaus.com
Author of Camel in Action: http://www.manning.com/ibsen
Make your Camel applications look hawt, try: http://hawt.io
Re: performance Replicated LevelDB vs. SAN
Posted by Oleg Dulin <ol...@gmail.com>.
I am running a similar test.
Replicated LevelDB, 3 Zookepers, 3 AMQ brokers with local_mem sync, I
am publishing messages on a queue using one thread, and taking them off
that queue on another thread.
Performance is abysmal, 1700 messages or so go out pretty quick, but
then it pauses every 50 messages or so.
The settings are mostly taken from the sample Replicated LevelDB configuration.
What gives ?
Any input is greatly appreciated.
On 2014-02-14 20:26:41 +0000, shippers said:
> I'm testing with ActiveMQ Replicated too (latest SNAPSHOT build) and seeing
> similar problems with stability and reliability. Performance-wise, it
> depends on the messaging pattern, number of pubs / subs, message size, and
> transaction size. Also, very dependent on network speed and local disk
> speed.
>
> Can you describe your testing env? Size of messages? # of clients pub and
> sub? Network speed?
>
> I'm running a 3 node ActiveMQ quorum_mem and 3 node Zookeeper.
>
> Example test, 1 pub and 1 sub client running on different hardware, 4k
> message size, persisted messages, not using transactions, 10G network,
> virtual hardware box of 2 CPUs and 8 GIGs memory.
>
> ~900 msgs/per sec steady state (pub / sub running normally, keeping up with
> each other)
--
Regards,
Oleg Dulin
http://www.olegdulin.com
Re: performance Replicated LevelDB vs. SAN
Posted by shippers <cv...@yahoo.com>.
I'm testing with ActiveMQ Replicated too (latest SNAPSHOT build) and seeing
similar problems with stability and reliability. Performance-wise, it
depends on the messaging pattern, number of pubs / subs, message size, and
transaction size. Also, very dependent on network speed and local disk
speed.
Can you describe your testing env? Size of messages? # of clients pub and
sub? Network speed?
I'm running a 3 node ActiveMQ quorum_mem and 3 node Zookeeper.
Example test, 1 pub and 1 sub client running on different hardware, 4k
message size, persisted messages, not using transactions, 10G network,
virtual hardware box of 2 CPUs and 8 GIGs memory.
~900 msgs/per sec steady state (pub / sub running normally, keeping up with
each other)
--
View this message in context: http://activemq.2283324.n4.nabble.com/performance-Replicated-LevelDB-vs-SAN-tp4677811p4677995.html
Sent from the ActiveMQ - User mailing list archive at Nabble.com.