Client has a requirement of additional SATA storage.
I came across this statement "Mixing disk shelves containing SAS disk drives and disk shelves containing SATA disk drives is supported, but not recommended" from DS4243, DS2246, DS4486, and DS4246 Disk Shelf Installation and Service Guide.
I could not find out such statement in latest version of this guide.
Regarding to performance, mixing disk types in SAS stack is not matter or minor factor. I/O transaction at SCSI level is asynchronous(bus is freed between request and response), so bus busy rate depends on traffic - not disk speed. On the other hand, disk speed may affect in FC-AL as packets are passed between disks in same loop so load of each disk may affect to overall latency.
Whereas, from perspective of reliability, SATA is less reliable so you should isolate both. SAS expander will be reset in communication failure. In rare case, all SAS expander in same stack or HBA will be reset by abnormal disk behavior.
The recommendation refers to the backend disk loops that connect the shelves to the controllers. You *can* put a SAS shelf and a MSATA/NL-SAS shelf in the same loop from the controller. This will affect performance as a whole (SAS would be slowed, MSATA/NL-SAS doesn't get faster) with an appropriate workload as the slower disks responses *could* limit the ability to drive and work with the faster disk responses.
Best practice is for each disk type (not necessarily shelf type, but it usually boils down to shelf type as well) to be in a separate disk loop to the controllers. For instance, SSD should be in a different loop than high performance SAS even though they might both be on DS2246 shelves, again because of the different IO performance nature of the disks.
In practice, I will mix on a single loop disks of similar IO characteristics but different capacities - for instance 600GB/1.2TB 10000RPM disks might be on the same loop. I tend not to do this by default, but will if time/space/port require. This type of concern comes into play usually as total controller ports are near being exhausted and I don't have freedom to power down and add port cards. It also highlights why pre-planning on controller deployment is important, as is the ability to predict the future 2-3 years out. Anyone who mastered that one please let me know.