At the moment, you’re probably best off with PCQ and Queue Tree.
Create a PCQ for your server to reserve the bandwidth, then create a separate PCQ for your workstations.
In the Queue Tree, create 2 parent queues (upload & download), then create 2 children for each (server & workstation).
What will happen, is that your server will always have bandwidth available and be able to use the full pipe when possible. Your workstations will equally share all bandwidth, up to the maximum of the pipe.
To ensure that this works as well as possible, I would suggest setting the parent queue bandwidth to about 95% of the actual available bandwidth (19m/1900k), and set your limit-at to about 75% of (15M/1500k).
I’ve been looking for a way to do this with simple queues, so I can have the Queue Tree available for QOS, but so far, I’ve not had any luck. In my case, QOS will just have to wait.
Good luck!