Opening of Faceplates slow

800xA v6.0.2

Client is virtualized Windows 10 LTSB2015 64bit

CPU 2.4GHz 2vCPU (CPU average <20% with 2 AfwWorkplaceApplication 32bit, 1 per display)

HDD is SAN (HPE Storevirtual) Tier 0 SSD MixedUse Raid 1, Tier 1 10k SAS Raid5

----------------------------------------------------------------------------

When I click link to Faceplate, the average time to open is ~3s, compared to thick clients where time is <1s.

When client have been running for som time (weeks) opening times may go up to 5-12s.

This also occures when multiple faceplates is open (pinned). Times increase linearly for 4+ opened faceplates.

The CPU average is at this point is 40% with 60% peaks.

----------------------------------------------------------------------------

Is there some way to identify where lag is coming from. Does not appear to be Network/CPU/HDD -related.

Comments (0) New comment

Fredrik Rank: 543

Asked 2 years ago

Answer this question Follow

Answers

The virtual solution lack hardware acceleration. Every pixel must be rendered by the workplace process (or rather its ESXi host processor) instead of a dedicated video accelerator card on a local and stunning fast PCI bus.

The virtual solution is dependant on video compression and decompression tied together over Ethernet. Some thin clients run on inferior, low energy/power processors that can not be compared with what you can put in a rich client.

The virtual solution mentioned run on a 2.4 GHz processor (multiple cores does not help since the workplace was not built for multicore architecture) which in relative terms, is rather slow.

I believe that if you were to add a thick client with a local SSD disk and a moderate graphics accelerator card, as comparison, it would run the very same workplace a lot leaner than the 20-60% CPU you mention, and hence be faster than the virtual. Add to that the fact that most thick computers can be equipped with processors beyond 4 GHz, without costing a fortune.

ABB recommend using regular computers (thick clients) for critical process operation and the most rich operator experience. The operator workplace will benefit from a fast processor, the more GHz the merrier.

Virtualized clients are supported, but will due to the above not run the System 800xA HMI as good as rich clients can do.

Comments (3) New comment

by Fredrik Rank: 543 on 8/31/2017 2:53:27 AM | Like (0) | Report

Thanks for your response Stefan.

Before you mention it, I know our VDI solution isn't supported by ABB. Our hope is still to get a good performing solution that might be supported when all kinks have been worked out.

I was under the impression that Workplace Application was unable to use hardware acceleration for graphics.
If this is possible we could add hardware acceleration in the form of Nvida Tesla GPU's to VM and reduce CPU load generated by Workplace Application.
But I do not see any significant CPU load by the Workplace Application. One Workplace gives about 20% load and 4 screen workplace + 3 Faceplates gives 40% load.
At this point graphics use <4% CPU. This is for VMware Blast to a Horizon Client but the figures is the same for PCoIP to a DELL P45 Zero-client.

Another option we have considered might be using a Teradici Apex2800 ASIC to offload the graphics from CPU, but with only 4% generated there is not much to reduce.

Do you have any experience with anyone using either Nvida Tesla/Grid or Teradici Apex2800?

We have previously been running thick clients with not much more than 2,4 GHz without problem so then shouldn't a offloaded CPU at 2,4 GHz be enough to run Workplace Application with good performance?

We have retained at least one thick client per controllroom when we upgraded to 800xA v6, and these client do a good job. Cost of clients is not the only factor, one must also count power and maintance costs.
VDI however offers a level of flexibility and hardware abstraction that is far supperior to thick clients and I'm realy hopefull that the lag can be minimized so that we can move 100% to VDI long term.

Also you mention that Workplace Application is unable to use multiple cores, and I must say I'm baffled by this as multi core applications have been standard for many years now.

by Stefan Stromqvist AKS expert

Rank: 1 on 8/31/2017 3:17:35 AM | Like (0) | Report

I believe you should file a support case or ask your regional ABB sales representative to assist you with connecting to our Product Management.

Version 6 is just one step away from Visual Basic (which was hatched when a processor only had one core). PG2 (running WPF and .NET) was added later on, but still the core design of the workplace is still "fitted" for what we had when VB6 was the only graphics.

The virtual scene is on a constant move; new technologies arrive every year - ABB can't be on every arena.

The level of hardware acceleration available for a workplace can be viewed in the PG2 Diagnostics Window (right click any PG2 window and select Diagnostics...) then select the Misc. tab and look for Render Tier.

The term Render Tier is explained on pages 352-353 of 3BSE049230-600B.

Level 0 = no acceleration.

Compare the same display between a rich and a thin session. The thin session will have Render Tier 0 unless accelerated - I cannot judge what the VDI and Nvidia Tesla can do here.

by Fredrik Rank: 543 on 8/31/2017 4:59:09 AM | Like (0) | Report

Thank you, I will definitely work with my sales rep. to get further support.

I've looked at render tier for some PG2's and all use tier 2 even though I currently have no hardware acceleration. My guess is that it's related to the VMware SVGA 3D virtual GPU.

Add new comment

Stefan Stromqvist AKS expert

Rank: 1

Answered 2 years ago

Question details

Opening of Faceplates slow

Answers

Other questions needing answers

Related topics

AKS experts