[TestReport] Comparing different PCs running blobfinders

Information and discussion regarding the ACE PackXpert software

[TestReport] Comparing different PCs running blobfinders

Postby TravisA » Mon May 05, 2008 10:44 am

May 2, 2008

    Purpose:
  • Experimentally determine the processing capabilities of various PCs in the lab by executing identical vision operations and recording the execution time of each.


    Machines Tested:
  • “Laptop” – Dell Latitude D810. Windows XP, 1.73GHz Pentium M processor, 2GB RAM, using an ADSTech PCMCIA 1394 adaptor. This machine has various other software installed and running (such as nfs server, RealVNC, VirusScan, etc)
  • “Matrox” – Matrox 4Sight M compact PC. Windows XP Embedded, 1.3GHz Celeron processor, 1GB RAM, built-in 1394b. This is a recently deployed PC, with very little other software installed or running
  • “Pentium4” – Generic Desktop PC. Windows XP, 3.0GHz Pentium4 processor (with HyperThreading), PCI 1394 adaptor. This machine has various other software installed and running (such as nfs server, RealVNC, VirusScan, etc)
  • “LittlePC” – AOpen mini PC MP965-D. Windows XP, 2.4GHz Pentium Core2Duo, 2GB RAM, built in 1394a. This is a recently deployed PC, with very little other software installed or running (but RealVNC server was running at the time)

    Hardware Setup:
  • 2 Basler 1394 cameras, mounted on a stand. The cameras are connected by 1394 FireWire to a Belkin 5-port FireWire hub. The hub is then connected by a FireWire cable to the PC being tested. The cameras, images, lighting, and results are all maintained between tests, so each machine is testing the same image. This way there should be no differences in image acquisition time. The only difference in hardware setup is the Matrox PC does not have a standard 1394a connector, so a 1394b to 1394a cable was used between the PC and the hub.

    Software Setup:
  • Each PC had Adept ACE version 1.5.4.9 installed with all of its underlying requirements: HexSight 3.4, DirectX 9.0c, .NET Framework 2.0, Sentinel Protection (license dongle)
    The same ACE Workspace was used on every system for consistency. The workspace has no controller or robot object. Each camera has an associated virtual camera, and a Blob Analyzer tool. A C# program object in the ACE workspace runs the blob tool 100 times, and records the start and end time of each operation. A second C# program runs the second blob tool, doing the same thing. A test C# program starts the previous two programs at basically the same time.

    Testing:
  • Each system was loaded, one at a time, and connected to the cameras. The test program was run, and all the data logged and stored. The cameras were kept in the same position, with the same lighting, and same objects in the field of view.
    Before testing, the Basler Pylon Viewer was used to ensure the camera settings were identical: in Acquisition settings, the Exposure Time is set to 500; in Transport Layer, the Packet Size is set to 1024.


    Results:
  • The variations in time were noticeable and significant enough to gain value from the test.
    The Matrox PC was by far the slowest, some 27% slower than the fastest PC on average.
    Somewhat surprisingly, the laptop managed to just edge out the Pentium4 desktop, beating it on average by 4%. The Pentium4 came in at 15% slower than the fastest. The laptop was about 10% slower than the winner.
    As the only dual core PC in the group, and being the newest PC, the AOpen littlePC was the fastest and most consistent.
avg times.jpg
avg times.jpg (77.14 KiB) Viewed 8394 times


execution time.jpg
execution time.jpg (173.07 KiB) Viewed 8391 times
TravisA
 

Return to ACE PackXpert

Who is online

Users browsing this forum: No registered users and 1 guest