How to route Gigabit Ethernet, remote PHY or remote RJ45 connector?

I need to design a 10/100/1000 BASE-T “backplane” board that contains multiple PHY outputs from Ethernet controller chips. On the board ,the LAN specifications are very strict in regard to how the LAN silicon(PHY) is situated to the magnetics and the RJ-45 connectors. My questions are primarily how to interface to an off the board RJ-45 patch panel that could be anywhere from ~2” to 28” away. Various forums have “addressed” this in the past, such as https://electronics.stackexchange.c...onnector-is-it-better-to-remote-phy-from-magn, “For remoting 1000Base-T Ethernet connector, is it better to remote PHY from magnetics or magnetics from connector?”. One opinion is to place magnetics close to the PHY and remote the connector. Another opinion is to put the magnetics close to the connector, remoting the PHY.
Which of the two, remote the connector or remote the PHY, is “correct”/ easier to implement for maintaining signal integrity and minimizing EMI?
What is the maximum length that is possible to remote the PHY or the connector?
Do I have to run the long lengths off the board?
How does a 48 or 96 port gigabit switch run their signals while maintaining signal integrity?
Are there any definite specs on how to proceed? Thanks in advance.

Apart from it's DC bias function, the ethernet magnetics mainly provides common mode isolation. Both sides are 100 ohms differential transmission lines. Impedance matching wise, the magnetics are almost transparent. There are two simple conclusions:

- the on-board part should be good 100 ohm matched differential pairs. The overall losses aren't affected much by the magnetics placing.
- isolation of external common mode interferences suggest to place the magnetics near to the PCB border respectively the connectors.

There are possibly contradicting points, e.g. suppresion of PHY born EMI. But on-board interferences can reduced by using high quality parts, off-board interferences can't be controlled essentially and it's difficult to prevent them from coupling to the ground plane and adjacent signals.