Developing Specifications for SonarWiz on AWS

Chesapeake Times, Vol 3 | October 2020

SonarWiz 7.7 will be the first release of SonarWiz to be officially supported on Amazon AWS. As we prepare to release this new version of the program, we have been working to develop best practices and recommendations for setting up an AWS environment that will run SonarWiz as efficiently as possible. We developed a short list of typical sidescan operations that we are using to benchmark different AWS configurations.

In our initial experiments we tested 13 different AWS configurations and one laptop for comparison and measured how long they took to complete our benchmark tasks in SonarWiz. The results are shown in Figure 1.
Figure 1: SonarWiz benchmark results for each AWS instance. Lower times represent better performance. The Lenovo T470p laptop result is indicated in red. 
There is both good news and bad news in these results. The bad news is, that even monster machines like the winning g4dn.16xlarge with 64 vCPUs and 256 GiB of RAM only achieved about 1.8x the performance running SonarWiz that we can get from a ThinkPad laptop. This has to do with a number of performance problems that occur when running desktop applications like SonarWiz on cloud platforms that were designed to share resources across web servers.

The good news is that you do not need to rent the most expensive machines available on AWS to achieve near-best performance. This latter point can be better seen if we plot benchmark performance versus the rental cost of the instance as in Figure 2.
Figure 2 Cost per hour to rent the instance versus total time to run the benchmark.

Machines in the lower-left corner of Figure 2 offer good performance running SonarWiz and low rental cost. In fact, instance g4dn.4xlarge achieved a benchmark time only 1.04x slower than the winner but cost only $1.94 per hour to rent and you can get pretty good performance out of machines that only cost 50 cents per hour. There is no reason to rent big iron if SonarWiz cannot yet fully take advantage of the additional resources provided by these huge servers. We found the sweet spot was a GPU-equipped machine with 16 to 32 vCPUs and 64 GiB of RAM.

All but three machines we tested were faster than the laptop we include for reference, but not by more than 2x. More than performance, the real win of running SonarWiz in the cloud is the flexibility it provides. Once you have created an image with SonarWiz and loaded your other mapping software, it is easy to log into that image from anywhere in the world, even from computers with vastly inferior power, like a Chromebook or the lab computer that hasn’t been replaced in years.

Cloud computing is the future of the industry and with the release of Version 7.7, SonarWiz will start that journey.

-David Finlayson, Chief Scientist