Scalability and performance of Thingworx

[spatel-21]

Dear experts,

How would the performance of Thingworx be affected when it deals with huge amount of data and large number of transactions?

As per this blog performance of neo4j is hit once data size increases upto 50GB.

https://community.thingworx.com/community/developers/blog/2015/12/01/integrating-dse-cassandra​  so it advises to implement DSE.

When would system performance start getting a hit with DSE? How would system behave with say 250GB of data and 100M transactions per month ?

Is there any other way to support system in such cases?

Is there any test data or experimental results for such cases?

[Aanjan]

Siddharth, we have some metrics from our base load testing:

Http requests - Streams:

Platform Cores	Platform CPU%	DSE Cores @ 4 boxes	DSE CPU %	# Platform Requests/sec	# DSE Writes/sec
16	72	16	15	17800	17800
16	70	8	25	18400	18400
8	79	8	20	10,200	10200
8	80	4	22	8,900	8900
4	79	4	2.5	4280	4280

Http requests - Value Streams:    

Platform Cores	Platform CPU%	DSE Cores @ 4 boxes	DSE CPU %	# Platform Requests/sec	# DSE Writes/sec
16	35	16	38	8200	82000
16	24	8	55	7200	72000
8	58	8	13	5,300	53000
8	25	4	25	2,900	29000
4	60	4	18	2500	25000

WS Requests - Value Streams - Single property update:    

Platform Cores	Platform CPU%	DSE Cores @ 4 boxes	DSE CPU %	# Platform Requests/sec	# DSE Writes/sec
16	35	16	5	18600	18600
16	10	4	21	17900	17900
16	5	8	35	12700	12700
8	25	8	12	17000	17000
8	11	4	31	11900	11900
4	59	4	21	11000	11000

WS Requests - Value Streams - Batch property update:

Platform Cores	Platform CPU%	DSE Cores @ 4 boxes	DSE CPU %	# Platform Requests/sec	# DSE Writes/sec
16	35	16	45	14000	140000
16	10	8	62	11600	116000
16	9	4	58	10200	102000
8	11	8	25	10,800	108000
8	35	4	12	9,600	96000
4	61	4	10	7800	78000

[spatel-21]

Hi Aanjan,

Thanks for the information. It is highly detailed and very useful.

However there are few points that I need better understanding of

1) '# Platform Requests/Sec' -- do this imply number of read requests or combination of read and write requests?

2) In the above table RAM size is not mentioned. How does RAM size affect the above data ?

3) Is there any particular reason why platform is much better able to handle WS requests rather than HTTP requests?

4) Also its clear that core size severely affects the performance. Is there any document or matrix that explains how to choose server size?

[mhollenbach]

We typically test on AWS instances with anywhere from 15 to 30 GB of memory. The most common tests seem to be on m3.2xlarge instances, and there are additional tests I'm seeing performed on the c4.2xlarge instances.

We have a Compatibility Matrix available under our support references documents:http://support.ptc.com/WCMS/files/166015/en/ThingWorx_System_Requirements_and_Compatibility_Matrix.pdf

This goes over the minimum hardware considerations when choosing a server.

Meghan

[spatel-21]

Hi Meghan,

It seems I do not have authorization to view the document. Is there any other way I can view it?

- Siddharth

[Aanjan]

Siddharth, do you have access to the eSupport Portal? The same document should be available under the 'Reference Documents' section.

[spatel-21]

I do have access to eSupport portal, but not to Refernce Documents.

Though I've upgraded the account, I'm not able to view it.

[gouimette]

Hello Siddharth,

I sent you an email directly regarding how you should be able to access our eSupport portal.  If you haven't received it, please let me know.

Best,

Gates Ouimette

gouimette@PTC.com

[czheng1]

hi,Aanjan

   "Platform Requests/Sec" -- do this imply number of read requests or combination of read and write requests?The size of the write/read requests ？What is the unit？

[qngo]

Hi, and how about PostgreSQL ?

Where Should I Store My Thingworx Data?​

Thingworx Persistence Provider guidelines​

[spatel-21]

Hello,

Thanks for the reply.

I'm currently evaluating different options. I'm also exploring a hybrid model with multiple persistence providers. That can combine PostGres and DSE.

What sort of challenges might be there in such case ?

Can you share any case study or success story or any use case that you might have seen?