Most of the time I see it increasing on the ingress port when frames cannot be delivered, because the destination device is unavailable. (e.g. frames on the fly while the destination device goes away). So the counter wouldn't point to the port that goes away but to the port the traffic is coming from. But beside of that common reason, er_other_discard can count up for various other reasons (for example frame routing problems), as it's a "collective-counter". I would first check if ports went offline in fabriclog -s.