Matt Bethel, the manager of systems engineering at Merrick and Company, provided a good overview of computer performance issues in the data-intensive LIDAR application space at the ILMF conference in Denver. Merrick has benchmarked a lot of different approaches in their need to increase throughput and decrease processing time for faster results. A good portion of investment is on the hardware/software side in a LIDAR shop, but time to process also factors in greatly to bottom-line costs in this process-centric industry.
Among the improvements and choices outlined by Bethel are:
- 64-bit platforms – hardware/OS/design software all as 64-bit means that more RAM is available, more stable, more data can be read in at once, less memory management needed
- Multi-core machines provide a processing time savings of 80% vs. single core and they prefer the Intel Nehalem processors, with a minimum of 8 processors as performance starts to plateau after 8
- 8GB or more of RAM is ideal
- GPU processing (general purpose graphic processing units) can assign processing tasks to the graphics cards (like NVIDIA’s CUDA) and GPUs are running a lot faster than CPUs — at 12x more speed currently. Pushing processing to GPUs increase speed anywhere from 1.5x to 100x faster. The industry will be able to handle the ever-increasing data density problem by utilizing GPU processors
- Hard drives also play a role with their speed (SAS or iSCSI connections vs. Fiber that can take a consultant to cofigure)
- Data processing from Network to Local to Network saves about 80% of time vs. using just network drive or reading/writing only local, with no build-up of data on the local drive to bog things down.
- 10Gbps ethernet network lines save roughly 40% of time as compared to 1Gbps cables
- Operating systems also impact time, with XP taking 70% more time than Windows 7
To wrap up the discussion, Bethel mentioned the move toward the cloud for processing, and he sees good promise in this approach although he hasn’t tested this approach and doesn’t know of good cloud solutions for LIDAR at this time.
