Cyclone Register 360 Plus - Point Cloud Cleaning - Performance Issues

[Geo-Scanner]

Hello Laser Scanning Forum,

thats my first day in the Forum. I am glad to be here and looking forward to getting help or providing help if possible.

My current issue is the following:
I am not yet an andvanced user of Cylcone Register but got some first experiences. I am currently working on a project with about 2400 setups with an overall size of about 1.3 TB. It is split into 35 separate windows.

Now when I start to do some point cloud cleaning, after the start of the program the performance is fast. Then when i do the second time cleaning, it processes (the bar goes up top 100%) and then the program is stuck about 5 minutes. After that I can work again nominally until I do the next cleaning. Only after shutting down the program and restart it, the first cleaning is fast. The second one then again takes up to 5 minutes.

My machine has the follwoing items:
- AMD Ryzen 7 5800X 8-Core Processor 3.80 GHz
- 32 GB RAM
- Windows 11
- NVIDIA GeForce GTS 1660 Ti
- 18TB hard disk, with 13TB free space

Thanks for any ideas, same experiences or any help.

[smacl]

Hi Veit,

At a glance, your PC is under powered for this kind of work. The most obvious shortcoming is the RAM, where 32GB is likely to get exhausted very quickly with processing work on large point clouds. The 5800x with 8 cores and 16 threads is also a bit on the light side, as is the 1660 Ti GPU but I'd guess the 32GB RAM is your main bottle neck. Th 18TB hard disk will also be very slow and I'd tend to supplement that with a 4TB NVME drive for current work, copying back to the bigger driver afterwards. Others here would know better, but I think Cyclone benefits significantly from fast drives and much less so GPU.

[badam]

Hi Veit,

At a glance, your PC is under powered for this kind of work. The most obvious shortcoming is the RAM, where 32GB is likely to get exhausted very quickly with processing work on large point clouds. The 5800x with 8 cores and 16 threads is also a bit on the light side, as is the 1660 Ti GPU but I'd guess the 32GB RAM is your main bottle neck. Th 18TB hard disk will also be very slow and I'd tend to supplement that with a 4TB NVME drive for current work, copying back to the bigger driver afterwards. Others here would know better, but I think Cyclone benefits significantly from fast drives and much less so GPU.

It is not cpu nor ram issue, his machine is perfectly fine even overpowered to handle this project, cleaning is slow, not matter what you replace but usually it is iops/drive speed affect this inside register 360, Of course little more ram would not hurt but in case of cleaning it won't help a thing.

But to be prepared cleaning should be done by setup for performance and result, or if it is not possible or not viable to do by setup for example large exterior point cloud, then you should do it with smaller chunks, like 100-200 setup then merge these together later...

Which version do you use? Since 2020.1 there are improvements to handle large scale projects but still lagging behind. Only issue with the old version it only suppports up to ~2000 setup (2048 opened filehandle).

my current pointcloud is gonna be 3600 setup. working from a ryzen 3900x, 128gb ram, over NAS and remote desktop. So everything which leica not recommend.

[smacl]

It is not cpu nor ram issue, his machine is perfectly fine even overpowered to handle this project, cleaning is slow, not matter what you replace but usually it is iops/drive speed affect this inside register 360, Of course little more ram would not hurt but in case of cleaning it won't help a thing.

If drive speed is an issue, an 18TB spinning hard disk would be a major bottleneck. Best easy thing to do would be fire up task manager while the processing is ongoing and see which resource is maxing out. The description below, where processing starts fast and then slows down, sounds very much like a resource being exhausted. While you say RAM isn't an issue, 32GB gets choked up very quickly on windows 11. Looking at the PC in front of me, with Visual Studio, Chrome and my mail client as the only foreground apps running, I'm seeing 21.4GB of RAM consumed. It takes very little in terms of point cloud processing to start hitting the swap file and grinding to a halt in this scenario. That said, only way to know for sure is to measure the performance and task manager is a first easy step.

Now when I start to do some point cloud cleaning, after the start of the program the performance is fast. Then when i do the second time cleaning, it processes (the bar goes up top 100%) and then the program is stuck about 5 minutes. After that I can work again nominally until I do the next cleaning. Only after shutting down the program and restart it, the first cleaning is fast. The second one then again takes up to 5 minutes.

[Carbix]

The fact is he needs to break up the linked groups. I know it's nice to clean all at once but at some point, it does not make sense to keep it all in one bundle.

Why clean a whole tower at once when you can clean floor by floor...

[gsisman]

Hello Laser Scanning Forum,

thats my first day in the Forum. I am glad to be here and looking forward to getting help or providing help if possible.

My current issue is the following:
I am not yet an andvanced user of Cylcone Register but got some first experiences. I am currently working on a project with about 2400 setups with an overall size of about 1.3 TB. It is split into 35 separate windows.

Now when I start to do some point cloud cleaning, after the start of the program the performance is fast. Then when i do the second time cleaning, it processes (the bar goes up top 100%) and then the program is stuck about 5 minutes. After that I can work again nominally until I do the next cleaning. Only after shutting down the program and restart it, the first cleaning is fast. The second one then again takes up to 5 minutes.

My machine has the follwoing items:
- AMD Ryzen 7 5800X 8-Core Processor 3.80 GHz
- 32 GB RAM
- Windows 11
- NVIDIA GeForce GTS 1660 Ti
- 18TB hard disk, with 13TB free space

Thanks for any ideas, same experiences or any help.

Veit,


I would agree with the Hard-Drive being an issue probably. You definitely see big gains with NVME & M2 drives, but as BADAM says cleaning with Leica is a slow process. Some of it is the logistics of their programming, some may be the way they write to the SQL LITE DB. It is faster cleaning one cloud at a time, but with an outdoor dynamic scene cleaning all overlapping scans in a section in one shot-though slower seems for us more efficient. I have notice a bit faster cleaning when running locally on the NVME drives than on our Synology RAID 10 NAS. Just remember with Register 360 you aren't deleting any points OUT of the database, you are just "flagging" them to hide them. We used Cyclone Register a little and maybe that database is the same, but if I recall you can delete delta tables by compressions and thus remove the points from the DB...not sure about this. I do believe it uses GPU RAM more in the cleaning process also as we had seen significant gains going from cleaning with 64GB RAM and a NVIDIA K4200 Card to 64GB RAM and a NVIDIA A5000 card

[Geo-Scanner]

Hi,

thanks a lot for all the responses.

I actually do the cleaning in the setup and not in the bundle cloud. So I click on a setup, go to setup view and choose the top view. Then I mark the desired aread with the fencing tool and press visual and cloud-cloud points cleaning. Or is this the wrong workflow? Some overviews are bundeled in one bundle (currently with about 600 setups in). The rest is in smaller bundels.

It is always and in each bundle the same behaviour. First cleaning perfectly fast, second (and third ....) extremly slow, 5 minutes more or less constantly. It seems to me as if the program is waiting for something. The task manager shows 0% activity in all areas during these 5 minutes. It seems to me it is just waiting for the "I am finished now" message. One fact supporting this is that when I start the cleaning process ( a second or third try) and shut down the program after a short time. E.g. 10 secs, restart the program and the project .... all points selected are cleaned up. So they are not visible anymore. So the process works actually fast .... but it just does not come to an end.

About the hardware facts, thanks for the ideas. I have the machine I have ... not much option to change it now during the project. However, I have smaller hard disk (4tb) as well, so will clean this up and use this for the project. As Temp Archive, I will the use the 18TB one.

Thanks!

[Carbix]

You're running out of memory probably both VRAM and RAM... You need to monitor that.

[smacl]

The task manager shows 0% activity in all areas during these 5 minutes.

You're running out of memory probably both VRAM and RAM... You need to monitor that.

@Geo-Scanner, maybe do a screen capture of each of the CPU,Memory,Disk & GPU sections of task manager and post them up here. 0% on everything is odd, for an 8 core 16 thread processor, CPU at 6.25% can mean that a single threaded process is running at 100% which is very common. If everything is at 0%, it is also possible that a message window can inadvertently get opened but hidden behind another window, try pressing alt+tab to cycle through windows to check for this. Other things that can really mess with performance are virus scanners, which can take silly amounts of time checking out each file your point cloud processing software writes to disk. If you have any security software installed, make sure to add exclusions for any disk locations used for point cloud and temp data.

If you're allowed even a minor upgrade to your PC, a 4TB nvme would be a very cheap upgrade that is going to be massively faster than an 18TB hard drive. Could be you existing 4TB drive is solid state and should be used.

[badam]

The task manager shows 0% activity in all areas during these 5 minutes.

You're running out of memory probably both VRAM and RAM... You need to monitor that.

@Geo-Scanner, maybe do a screen capture of each of the CPU,Memory,Disk & GPU sections of task manager and post them up here. 0% on everything is odd, for an 8 core 16 thread processor, CPU at 6.25% can mean that a single threaded process is running at 100% which is very common. If everything is at 0%, it is also possible that a message window can inadvertently get opened but hidden behind another window, try pressing alt+tab to cycle through windows to check for this. Other things that can really mess with performance are virus scanners, which can take silly amounts of time checking out each file your point cloud processing software writes to disk. If you have any security software installed, make sure to add exclusions for any disk locations used for point cloud and temp data.

If you're allowed even a minor upgrade to your PC, a 4TB nvme would be a very cheap upgrade that is going to be massively faster than an 18TB hard drive. Could be you existing 4TB drive is solid state and should be used.

I can assure you that it is staying around zero all the time. I remembered this year when i had to add more clouds to our biggest project (7854 setup) and i was forced to reopen everytime i deleted some points, it's cause is related to setup count. not hardware rather then software issue.

If i was fast enough to delete point i could delete maybe twice in one open.


Solution is to duplicate the project slice it up to 2 part then clean then remerge the project and register everything together.

In register 360 there are times when it's just sits there with literally no activity. no cpu no disk or network. One of this example is sitemap change with big amount of setups 1500+setup.