Community
Select to view content in your preferred language

Understanding Workspace Throughput in ArcGIS Reality Studio

733
0
04-01-2025 04:02 AM
EmreÖzdemir
by Esri Contributor
Esri Contributor
5 0 733
‎04-01-2025 04:02 AM

Have you ever wondered how fast you can process your projects? Are you concerned about meeting your targeted delivery deadlines? Understanding the throughput of your workspace is crucial for effective project planning and setting realistic expectations. In this article, we will guide you through computing workspace throughput and share some production rates we achieve with our hardware.

 

Evaluating Performance with Workspace Throughput

In this section, we will introduce workspace throughput, a metric we use to measure the processing performance of ArcGIS Reality Studio projects.
 

Understanding and Measuring Workspace Throughput

Workspace throughput is the number of pixels processed in a workspace within a day. To calculate it, you need two values: project volume and computation time.

  1. Project Volume: This is the total pixel count, measured in gigapixels, considering all input images. It is calculated by multiplying the number of images by their corresponding resolution. For example (using information from the screenshot below):
      • Capture Session:
        • 300 nadir images with 492.8 Megapixels (35,000 x 14,080) per image.
        • 1048 (246+248+248+306) oblique images with 151.3 Megapixel (10,652 x 14,204) per image.
      • Calculation:
        • (300 x 492.8 ) + (1048 x 151.3) = 306,403.456 Megapixels
        • 306,403.456 Megapixels / 1,000 = 306.4 GPix (Volume of the Capture Session)
    Number of images and image resolutions can be seen in Project Tree and Capture Session.Number of images and image resolutions can be seen in Project Tree and Capture Session.
  2. Elapsed Time: This is the time, in days, required to complete the reconstruction. ArcGIS Reality Studio displays this information directly in the Process Manager. For example:
      • Example (from the screenshot below): 3.20 days and 40 minutes ≈ 3.23 days
    The elapsed time of a reconstruction can be found directly from the Process Manager.The elapsed time of a reconstruction can be found directly from the Process Manager.
  3. Workspace Throughput Calculation: Divide the project volume by the elapsed time.
    • Calculation: 306.4 GPix / 3.23 days ≈ 94.9 GPix / day
    • As we had two workstations in this sample workspace:
      • 94.9 / 2 ≈ 47.5 GPix per day per workstation.

Real-World Numbers from our Office Workspace

In this section, we share an overview of the throughput values we achieved in our office workspace for recent projects. The performance of your Reality Studio workspace varies based on several factors, including disk speed, CPU/GPU memory, image size, image overlaps, scene complexity, network bandwidth of connected computers in a distributed processing cluster, and the chosen products and formats.

In our projects, we use workstations and Network Attached Storage (NAS) units with our recommended hardware specifications. Below, we summarize these hardware specifications. Detailed recommendations can be found in our public documentation.

Additionally, to optimize performance, we store our datasets (input images) and project workspace (outputs) on separate NAS units.

The numbers below are specific to our setup and are intended to help you gauge potential performance. We recommend generating your own benchmark values as explained above, using large datasets (>1TPix of input images) for reliable measurement. These benchmarks can assist in future project planning, such as estimating the number of workstations needed to complete a project within a given timeframe.
 

Alignment Throughput

As the alignment step of Reality Studio runs on one workstation, you can easily compute the throughput of the alignment workflow by using the volume of the imagery in relation to the time it takes to complete the alignment process.

The alignment throughput we achieve is approximately 5 TPix per day.

Reconstruction Throughputs

Below, we provide an overview of the throughput values from two different reconstruction projects in our workspace, where we selected default products for each scenario:

Quality Ultra, Scenario Aerial Oblique
  • Project Volume: Greater than 1 TPix
  • Image Overlaps: Standard overlaps of 80% forwards 60% sidewards
  • Workspace Throughput: Approximately 60 GPix per day per workstation.
Quality Ultra, Scenario Aerial Nadir
  • Project Volume: 2 TPix
  • Image Overlaps: Standard overlaps of 80% forwards 60% sidewards
  • Workspace Throughput: Approximately 85 GPix per day per workstation.

Office Workspace Setup

Below is a sample configuration of our workstations:
  • Processor: AMD Ryzen Threadripper 3960X 24 cores
  • Memory: 64 GB RAM
  • Graphics Card: NVIDIA GeForce RTX 3090
  • Storage: Processing directory on SSD
  • Network: 1Gbps Ethernet connection with a stable network
  • Software: ArcGIS Reality Studio 2024.3.2

Conclusion

With some simple calculations, you can effectively plan your projects for your processing cluster. Adhering to the recommended hardware specifications allows for significant production throughput.

We hope this overview provides valuable insights into the capabilities of our software and assists you in your future projects. Significant deviations from our numbers may be due to network or hardware bottlenecks. Ensure to run a large dataset for accurate benchmarking. If issues persist, please contact our support team.

We appreciate your feedback and questions, so please share them in the comments below. Stay tuned for more updates and enhancements as we continue to evolve ArcGIS Reality Studio to meet your needs. Thank you for being part of our community!

Contributors