No Performance Benefit increasing CPUs or Disk I/O on VM Servers?

PIX4Dmapper
1

I was monitoring the CPU (16 Xeon VM), GPU (Tesla M60), Memory (164 GB), and EBS gp3 (3000 iops, 125 MiB/s throughput) on my AWS EC2 Windows Server. The processing compared to other software I’ve used, local and SAAS, seems relatively slow in Pix4D. I noticed that Disk I/O rarely approached 100 MB/s though. Memory usage was always below 30%, CPU jumped all over the place depending on the stage of processing but was only pegged at 100% for a minority of the time (can’t really say what percentage - guessing maybe 20%, 30% at most?). And GPU was hardly used. When it wasn’t idle at 1% it rarely went above 20% usage and like the CPU for a minority of the processing.

So I changed the EBS gp3 storage from 3000 iops to 16,000 ips and throughput too 500 MiB/S. No significant difference in processing time.

So now I’m embarking on trying to figure out if there’s any bottleneck to be alleviating. It’s definitely not available RAM. Disk I/O doesn’t seem to the the bottleneck nor GPU. But if CPU is only maxed 20% of the time, that can’t be a big bottleneck. Is there a point where the performance increase from adding more computing resources is next to nothing?

(These are tests doing multispectral image processing with typically around 10,000 total images each 2.5MB in size.)

2

Hi jt1,
It is often very difficult for us to recommend specific hardware setups as they continually changing. As a result we have teamed up with a company called Puget Systems which tests our software with different hardware configurations. I would suggest visiting their site and direct some specific questions to them.