more quick actions
minimize quick actions
Hello all,
i got a customer that he wanted to know our HDS performance.In his project, they need to record more than 1000 registers every second, and some of them(200-300) every 100ms.
i told him it's ok for PcVue, but he wanted us to show him the HDS performance in document.
I found nothing in kb on this topic.:(
thanks a lot!
Posted : 06/08/2012 10:29 am
Hi, you have to ask to Arnaud Degré, I think that he did some test.
But I think you are very optmistic with 1000 records per second + 200 or 300 records each 100ms it makes 2000 or 3000 records each second more.
I'm not sure that HDS can support this flow of records.
Posted : 06/08/2012 8:05 pm
Thank you,Romain. I'll refer to him.
Sad to know that you think i am too optmistic. 🙁
Posted : 07/08/2012 6:52 am
Hi Shuquan,
I did recent tests and it seems we could absorb about 3000 per seconds.
But, actually, what we had in reality was 10000 data changing every 3s which is an average of 3000 event/s
With this configuration we had 6000 trends always in the pending buffer but...no problem. Note that we had a very good computer (server, quad core, etc...)
I know that the tests document done by Arnaud was very less optimistic. I think it should be good to redo a platform
What protocol your client is using?
Posted : 07/08/2012 10:05 am
Great, Nico.
Could you share the documenet with me?
At the moment, they have not decided which protocol, but we know there are severial options: PVI from B&R, S7 Tcp/ip, OPC for AB PLC...
Posted : 07/08/2012 10:28 am
There is no official document...sorry
For your information those tests have been done using s7 tcp/ip protocol
Posted : 07/08/2012 11:04 am
Important point to keep in mind is that if you SQL Server you have to maintain your base and some SQL task block access to table(like defrag).
So when this king of tasks are executed HDS can't insert data and the HDS buffer can reach a very huge size.
If the Defrag task take 10min (600seconds), 3000 trends at 1s, one records is about 150bytes.
If I don't make a mistake: 600 * 3000 * 150 = 270 000 000 bytes
I don't know if HDS is able to empty 270Gb during the normal workings.
I hope I will be clear.
Posted : 07/08/2012 11:40 am
Yep good remark.
So the best is ....dont defrag! 🙂
Posted : 07/08/2012 4:19 pm
Usual limits that are told to customer is about 300 records/s with a continuous and stable system (always 300rec/s never les or more).
You can higher this limit with these recommendations:
- use fast hard drive
- locate log file on one physical hard drive, and data file on an other one
- use at least a quad core wit a x64 OS and a Standard SQL edition
@Nicolas: I'm surprised you achieve to reach 3000rec/s. I'm curious to have more information about your computer and SQL configuration
Posted : 13/08/2012 1:02 pm
:blink: only 300records/s? How can HDS deal with large projects?
Posted : 13/08/2012 1:41 pm
It's very hard, but there are some solutions.
First, there is simple recommandation: you can filter the datas before recording them (is it really mandatory to archive every data every ms ? what for these records will be used)
Then, trends can be recorded separately by different servers.
You can also record data in a specific file format and then import them into your database.
Note that PcVue record values only whene they change, but it's possible to resample values after in the db.
Posted : 16/08/2012 2:51 pm
By configuring dead band AND refresh period in trend groups.
Posted : 20/08/2012 2:16 pm
🙂 they are good solutions...
Posted : 21/08/2012 12:36 pm