Without knowing what you are simulating, it is difficult to point you in a direction. I will offer that a big mistake many people make is to jump directly into the big, bad, ugly, and 3D problem. It sounds like you may be making this mistake. Instead, there is likely a bunch of smaller simulations you can perform that will get your design very close to finished before ever having to simulate 3D. For example, if your device is using a photonic crystal with a band gap to guide light, you could first simulate a single unit cell and design it to give you the band gap you need. Second, you could perform a smaller simulation to see if the waveguide supports any modes. If your device has metamaterials, you can simulate those and extract the effective properties. You can then run 2D simulations using just the effective properties. Many times you will not even need to run the big, bad, ugly, and 3D problem. There are lots of tricks like this. Without knowing your device, I can only talk in generality and give a few examples.

Mesh size is an important parameter to simulation time, so i suggest that change it manually. Also, you can use the nonuniform mesh sizes for different regions of the device.

Thank you for your answer. The mesh sizes can be increased to a maximum of 1/10 to 1/15 of the wavelength. The sizes larger than this will cause inaccuracies to the solution. However, by considering the maximum mesh size, the problem is still very complicated to be solved using even a high-spec server.

Without knowing what you are simulating, it is difficult to point you in a direction. I will offer that a big mistake many people make is to jump directly into the big, bad, ugly, and 3D problem. It sounds like you may be making this mistake. Instead, there is likely a bunch of smaller simulations you can perform that will get your design very close to finished before ever having to simulate 3D. For example, if your device is using a photonic crystal with a band gap to guide light, you could first simulate a single unit cell and design it to give you the band gap you need. Second, you could perform a smaller simulation to see if the waveguide supports any modes. If your device has metamaterials, you can simulate those and extract the effective properties. You can then run 2D simulations using just the effective properties. Many times you will not even need to run the big, bad, ugly, and 3D problem. There are lots of tricks like this. Without knowing your device, I can only talk in generality and give a few examples.