On Fri, Aug 19, 2016 at 2:23 PM, Supun Sethunga <[email protected]> wrote:

> Hi all,
>
> Please find the deployment setup details at [1]. Some parts of it are yet
> to be completed though.
>
> @Imesh: No, we haven't created any puppet scripts for the setup.
>

​Thanks Supun! Few things I noticed:

   - ​
   ​As I see we have run this on AWS. If so we can create a Cloud Formation
   ​script to automate the entire deployment (may be Puppet can be internally
   used but not mandatory).
   - The reason why we need automation for this is, as I believe
   performance tests should be able to run by any user for verifying the
   results. We have done this in MSF4J.

   - Is there any reason to use a VM for MySQL without using RDS?

T
​hanks



> ​
> @MalithJ: Input was controlled using Jmeter. Throughput is measured at the
> receiver side, in databridge level. (We can get it using
> -DprofileReceiver=true property)
>
> @Sanjeew/Harsha: This was tested with the default protocols (Binary for
> throttle publisher and thrift for the other publishers), but with tuned up
> data-agent-configs. Here I used four separate standalone APIm nodes, only
> to generate the events. This was not possible to simulate with a thrift
> client like we did for ESB Analytics, because here we have multiple
> receivers (unlike in ESB Analytics case), and the events are received at
> differents rates to those different receivers. We can publish to the
> streams separately, but it would not be same as a real world scenario..
> Here the limitation of the throughput is caused by the complexity of the
> execution plans and the stream persisting, at the Analytics Server.
>
> [1] https://docs.google.com/document/d/1A1Pq-
> nOi9PmGdZz3wNovPc7kG3vc74p93tOTh_WYp_o/edit#
>
> Regards,
> Supun
>
> On Fri, Aug 19, 2016 at 1:55 PM, Sanjeewa Malalgoda <[email protected]>
> wrote:
>
>>
>>
>> On Tue, Aug 16, 2016 at 10:43 PM, Harsha Kumara <[email protected]> wrote:
>>
>>>
>>>
>>> On Tue, Aug 16, 2016 at 5:29 PM, Supun Sethunga <[email protected]> wrote:
>>>
>>>> Hi all,
>>>>
>>>> We carried out some performance tests for APIM Analytics server in a
>>>> Minimum HA setup. Idea was to saturate the Analytics server up to the
>>>> maximum number of requests it can handle, without APIM (client side)
>>>> getting their publisher queues full, see how it will behave in the long run
>>>> (for couple of hours). With this test, found out that the maximum
>>>> throughput it can handle is around 18,000 requests / sec. Please see below.
>>>> For this test, used tomcat echo service as the backend. Used 4 standalone
>>>> APIM servers to achieve the throughput.
>>>>
>>> This means if we scale up the API gateways, it will lead to getting
>>> queue fill messages frequently when request throughput to DAS node is
>>> larger than 19000 TPS. Since DAS node can handle more than 19000 TPS, this
>>> behavior may cause due to rate of DAS server accepting messages is less
>>> than the rate of message coming to gateways which leads to filling the
>>> queues in gateways. Is this tested with binary or thrift transport? May be
>>> better if we can have results for both binary and thrift transports.
>>>
>> When we test API Manager throttling we sent throttling events to CEP data
>> receiver, we were able to send much more requests than this.
>> I think in DAS also we are using same data receiver implementation(like
>> harsha mentioned is it on thrift or binary?).
>> If i remember correctly single data receiver handled 30000 or more
>> requests(we had 10 gateways or so).
>> I think here we need to check how data receiver and analyzer behaves when
>> it have high load. So we may directly test it with client tool. If you test
>> with gateway then anyway it will fail at 4500TPS(even if Que filling issue
>> sorted)
>>
>> Thanks,
>> sanjeewa.
>>
>>>
>>>> Raw data can be found at [1].
>>>>
>>>> [image: Inline image 1]
>>>> If we increase the throughput beyond 19,000 requests/sec, then the
>>>> queues at APIM side get filled after sometime. Please refer the 2nd, 3rd,
>>>> and 4th sheets of [1]. The huge variance towards the end in those charts
>>>> were due the queues getting filled.
>>>>
>>>> I will add these to a performance results template, including the
>>>> advanced details about the setups and the configs.
>>>>
>>>> [1] https://docs.google.com/spreadsheets/d/1vVRPp8b31DccJ1A6
>>>> p9MRGxzmLAX2yiGAebT25kpsLzg/edit#gid=0
>>>>
>>>> Regards,
>>>> Supun
>>>>
>>>> --
>>>> *Supun Sethunga*
>>>> Senior Software Engineer
>>>> WSO2, Inc.
>>>> http://wso2.com/
>>>> lean | enterprise | middleware
>>>> Mobile : +94 716546324
>>>> Blog: http://supunsetunga.blogspot.com
>>>>
>>>
>>>
>>>
>>> --
>>> Harsha Kumara
>>> Software Engineer, WSO2 Inc.
>>> Mobile: +94775505618
>>> Blog:harshcreationz.blogspot.com
>>>
>>> _______________________________________________
>>> Architecture mailing list
>>> [email protected]
>>> https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture
>>>
>>>
>>
>>
>> --
>>
>> *Sanjeewa Malalgoda*
>> WSO2 Inc.
>> Mobile : +94713068779
>>
>> <http://sanjeewamalalgoda.blogspot.com/>blog
>> :http://sanjeewamalalgoda.blogspot.com/
>> <http://sanjeewamalalgoda.blogspot.com/>
>>
>>
>>
>
>
> --
> *Supun Sethunga*
> Senior Software Engineer
> WSO2, Inc.
> http://wso2.com/
> lean | enterprise | middleware
> Mobile : +94 716546324
> Blog: http://supunsetunga.blogspot.com
>



-- 
*Imesh Gunaratne*
Software Architect
WSO2 Inc: http://wso2.com
T: +94 11 214 5345 M: +94 77 374 2057
W: https://medium.com/@imesh TW: @imesh
lean. enterprise. middleware
_______________________________________________
Architecture mailing list
[email protected]
https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture

Reply via email to