Do involve the project stakeholders as early in the process as possible.
This should include the architect, owner, fire protection engineer/life safety consultant, mechanical/electrical engineer, and contractor. It is important for the mechanical engineer and fire protection engineer to work in concert. Perhaps most importantly, the authority having jurisdiction (AHJ) should be involved at the appropriate time, especially if computer modeling will be utilized. Presenting a Fire Engineering Design Brief that summarizes the proposed smoke control approach, types of modeling, number/nature of design fire scenarios, expected model outputs, and pass/fail criteria can go a long way in establishing a positive relationship and avoiding surprise requirements/requests after all the modeling has been completed.

Do choose a smoke control approach appropriate for the application.
This author was retained during construction of a project where the contractor was attempting to achieve a pressure differential between two large-volume spaces. The pressurization method was the wrong approach for this building, and achieving the pressures proved unfeasible. One should involve a fire protection engineer or smoke-control-savvy mechanical engineer and create the right game plan from the beginning.

Do not spend a lot of time creating complex computer models too early in the design process.
The design will change. Programming will be updated. Horizontal openings will move, grow, and multiply. One should hold on the computer modeling until the schematic design is finalized. A relatively stable BIM model can be used to import the geometry into a fire modeling program instead of building it painstakingly by hand.

Do use defendable and appropriate design fire scenarios.
National Fire Protection Association (NFPA) 101, Life Safety Code, lists eight scenarios that are required to be considered for performance-based design. Along with design fire locations, the maximum size and characteristics of the design fire are important. Just because fire test data may be available for a burning chair, is that a severe-case fire for the space in question? Documentation of these assumptions and inputs tie in to the aforementioned Fire Engineering Design Brief to be presented to the AHJ early on for its buy-in.

Do not put all the exhaust in one place.
Even if the exhaust volume satisfies the hand calculations, providing all 200,000 cfm in one corner of the atrium will probably not be effective for a design fire in distant locations. Firstly, concentrated exhaust can create plugholing, where the exhaust is so strong it pulls fresh air in through the smoke layer and therefore renders the system useless. Secondly, if smoke is required to be pulled long distances across a ceiling, it entrains more air and creates more smoke, so the original design exhaust volume may prove inadequate. Distribute exhaust inlet locations as broadly as possible to create an even exhaust scheme. Computational fluid dynamics (CFD) modeling illuminates these issues.

Do look and plan ahead for the testing and inspection process.
If possible engage the commissioning agent during the design phase to coordinate and communicate the design of the system, establishing a commissioning plan in the process. It is important to define not only what is being tested, but also the schedule, order, and the pass/fail criteria.