DIST sytem - node sizing

Find and share HowTos to various installations / configurations!
6 posts • Page 1 of 1
vbalan007
Posts:20
Joined: Fri Jul 28, 2017 7:11 am

DIST sytem - node sizing

Post by vbalan007 »

Hi,

How DIST system nodes been sized.
is it based on Based on Limitation of Event manager?
single threaded. can handle 100000 events (maxInputMsgCount).

Is there any way to capture and store the no. of events handled by the event manager?

What are the other factors to be considered?

leoknipp
Posts:2928
Joined: Tue Aug 24, 2010 7:28 pm

Re: DIST sytem - node sizing

Post by leoknipp »

The config entry maxInputMsgCount defines the maximum number of messages which can be queued (temporary) in the input message queue of the Event Manager for a given period of time (msgQueueLimitTimeout). If the time limit is exceeded a the number of message is still higher than the limit the connection to the specific manager will be closed.
For detailed information please have a look at the WinCC OA doucmentation: Installation --> Error tolerance and stability.

With the message statistics functionality you can count the message send/received by a specific manager. It is the number of messages and not the number of value changes. In a single messages several value changes can be send.
For detailed information please have a look at the WinCC OA documentation: System management --> Diagnostics --> WinCC OA diagnostics.

What is meant in detail with "How DIST system nodes been sized."?

Best Regards
Leopold Knipp
Senior Support Specialist

vbalan007
Posts:20
Joined: Fri Jul 28, 2017 7:11 am

Re: DIST sytem - node sizing

Post by vbalan007 »

Thanks a lot.

Here is the Scenario:


I have 2000 DPE from a driver which sends data at a rate of 100 ms. Which produce 20,000 value changes per second. Does this 20,000 value changes can be processed by Event manager (Driver--> EM--> DM--> VA /RDB) without any latency / buffer queue ?

Do we need to add nodes? (another EM thread). if so how do we size the nodes based on EM performance?

Thanks,
Bala

leoknipp
Posts:2928
Joined: Tue Aug 24, 2010 7:28 pm

Re: DIST sytem - node sizing

Post by leoknipp »

If the system is capable of handling a load of 20.000 VC/sec depends on several factors:
-- used hardware
-- operating system
.. configuration of the system (single server/redundant server)
-- complexity of the data model
...

To ensure that the chosen setup is the correct one you can make a load test with a configuration similar to the real project.

Normally if the PLC is sending values which need to be processed by the driver some of them need not to be send to the Event Manager, e.g. by using appropriate smoothing settings (old/new comparison for boolean values, value/time dependent smoothing for analog values).

How many value changes do you expect to be send to the Event Manager?

Best Regards
Leopold Knipp
Senior Support Specialist

vbalan007
Posts:20
Joined: Fri Jul 28, 2017 7:11 am

Re: DIST sytem - node sizing

Post by vbalan007 »

At this point, approximately 15000 values / sec after driver side smoothing. Does the Event manager of a single node has a capacity to handle these values with out latency / buffering?

As the EM is a single threaded, as per recommendation, we couldn't go beyond 4 core processor with highest clock speed.

if we configure for redundant pair, I don't think the two systems EM will share the load.

We can go for DIST nodes, but on what basis we could do the No. of DIST nodes sizing?

leoknipp
Posts:2928
Joined: Tue Aug 24, 2010 7:28 pm

Re: DIST sytem - node sizing

Post by leoknipp »

At NFR tests made by ETM we have done long term tests with a constant load higher than 15.000 VC/sec. These tests have been made in defined conditions (no additional tasks, limited amount of archived values, given amount of alerts) with a specific hardware.
As written in my previous post it is not possible to give an answer if your system is also capable of handling the load without detailed information.

You can do a test on the hardware which shall be used with the WinCC OA project and the load which is expected.
At this test you have also take into account
-- how many of the value changes need to archived
-- how many alerts will be generated
-- which type of archiving is used

After this test you can make a decision if the system needs to be splitted in several systems.
In case of splitting the systems you also have to take into account how data is handled afterwards and if you do not create too much additional load by sharing the data for all systems.


In a redundant system the load is not splitted. The active event manager has to send current data to the passive one which will increase the load.


If you are interested we can give you a NFR Benchmark test project (compatible with WinCC OA 3.15) which was made by ETM.
In that case please get in contact with the ETM support using the official Siemens Support Site:
https://support.industry.siemens.com/cs/start?lc=de-AT

When submitting a service request please follow the instructions described in an FAQ:
https://portal.etm.at/index.php?view=it ... &Itemid=54

Best Regards
Leopold Knipp
Senior Support Specialist

6 posts • Page 1 of 1