We were lucky and got a few new workstations with high core counts (Core i9-13900k) and while compiling one of our support projects (it compiles a bunch of dependencies) I noticed that the git clone (though https) slows down to about 1/100 of the usable bandwidth. At our building we have 100Mbps of internet bandwidth (no comments please). When we kick off the process of “make -j” there are about 10 processes that all do the “git clone” thing through an External_Project_Add() function. most of those processes will utilize the max bandwidth as much as possible. As other projects get done cloning, those large projects (ITK/VTK) will pick up the extra bandwidth. This is what everyone would expect. But not every process seems to understand that there is more bandwidth available and they stay stuck at about 1Mbps of bandwidth. I have repeated this on an older Linux workstation and a pair of the new workstations, one running Linux and one running Windows 11. I am able to monitor the bandwidth utilization of the entire building so that I know no-one else is hogging the bandwidth. This is the first I’ve ever really seen this happen. I just tried this again on my MacOS machine (Core i9-9900K) with “make -j16” and I get the same issues. This is all with relatively recent versions of CMake and Git and using “Unix Makefiles” as the generator type (Can’t use Ninja as there is a long standing HDF5 bug when using Ninja).
1009:[mjackson@Jolder:Test]% git --version
git version 2.24.3 (Apple Git-128)
1010:[mjackson@Jolder:Test]% cmake --version
cmake version 3.21.1
CMake suite maintained and supported by Kitware (kitware.com/cmake).
The project is a GitHub - BlueQuartzSoftware/NXSuperbuild: Dependent Libraries for complex and DREAM3DNX projects. Either this is new or I have just never noticed this before?
I’ll need to take a machine home to try from there and see if there is some odd internal issue.
Has anyone ever noticed anything like this before?