There are really great points which were mentioned in this thread (I mostly liked the first response by Thom A which can be great opening of a blog that summarize the topic), but I feel that it is a missing a bit more specific points like discussion about performance. Please do not see this message as an answer since it cannot stand by it self! This come only if and when the use of SQL Server on Linux actually fir your case. I just want to add to all that was already said.
We did a lot of performance tests back in 2017 and before during the private preview> public preview> and after the release. The test were done mostly on UBUNTU since this is what we have for mongoDB and the idea was to put all in one environment.
My 2 cent from the performance aspect is this:
We found clear performance advantage in using SQL Server on Linux in small environments which used (near)default configuration with no optimization. The overhead of the operating system took big part in this case and not just the SQL Server directly. Once we moved to bigger host systems the advantage moved slowly to the windows hosting, but in one point when we examine big data of tens of Terabytes it actually moved back to advantage for Linux.
Important! (1) SQL Server on Linux supports only ext4 (you will need to split your files to 16 Tbytes or less) and <span style="text-decoration: underline;">XFS file system which is the proffered according to our tests</span>. (2) Use VERY low value for the Swappiness (default is 60 and best is 10 or less)! (3) MongoDB is not "NUMA-aware" but SQL Server is designed to take advantage of NUMA - This put a challenge for us. Our conclusion was that Mongos play OK with SQL Server but Mongod should be in different machine (If must, possible to put the config server wit the SQL server and in this case execute mongod using "numactl --interleave=all"). In this architecture NUMA can stay enabled probably.
>>>> "Running MongoDB on a system with Non-Uniform Access Memory (NUMA) can cause a number of operational problems, including slow performance for periods of time and high system process usage."
... to be continue in some other day 😉 ...
You can check this document regarding the topic (It is mostly promoting and did not gave me a feeling of technical document with numbers but yet it does brings some real case scenario).