I’ve got an RB450G connected to 2 DSL isps via bridged modems one providing ~72/20 and one providing 14/1 (One is VDSL2 and the other is ADSL2+) both connections are PPPoE
Fairly basic Nat/Firewall setup, 1 Mangle rule to direct inbound traffic on the 2nd isp to another routing table (so the reply goes back out on the right line) and a couple rules to clamp the MSS on the PPP connections
The ADSL line doesn’t really get any traffic other than ICMP echos for the latency monitoring and SNMP polls from a remote libreNMS box, It does push a default route into the main table, but with a distance of 10 so it will only get used if the VDSL2 line loses the PPP session.
There is a bufferlo “Router” running in AP mode connected to one of the Ethernet ports that’s being used as a dumb 802.11ac (5Ghz) and 802.11n (2.4Ghz AP). Also the Ethernet ports are slaved/switched for the LAN side.
Interestingly I had some friends round the other week who were trying to play TF2, I’d not installed it on my laptop so was downloading it from steam and was causing them latency spikes of around 200ms+, it seemed the RB cpu was spiking to 90% at around the same time.
Now I know the RB450G is more powerful than the RB7xx series, however the RB7xx supports fasttrack so I’m wondering if I’d be better off using an RB7xx instead at home, since neither of my Lines are faster than 100Mbit/s I don’t really need the Gig interfaces at the moment.
I don’t really want to buy a new routerboard at the moment as Ideally I’d want to wait for either the 3011 or the hAP-pro to be available here.
Interestingly I had some friends round the other week who were trying to play TF2, I’d not installed it on my laptop so was downloading it from steam and was causing them latency spikes of around 200ms+, it seemed the RB cpu was spiking to 90% at around the same time.
Two different issues here:
CPU load: Maybe it has to do with your firewall/mangle rules
latency: do you have QoS set?
The only mangle rule I have other than the TCP MSS clamping is the one to direct traffic coming in on WAN2 to the 2nd routing table, that interface usually does about 1 - 10pps as other than incoming pings and SNMP polls from the NMS there shouldn’t be any traffic on it unless the primary goes down then it will get used.
QOS- I Don’t although the ISP for the VDSL line does do some kind of traffic prioritisation on the downstream see http://www.plus.net/support/broadband/speed_guide/traffic_management.shtml
If you want to be able to saturate your internet line without causing latency for online games you’ll have to implement QoS on your end.
The ISP already does downstream QOS, if their traffic management system is working properly it should treat real-time protocols such as gaming as higher priority.
doing it at the CPE for downstream isn’t that efficient since the only thing you can really do at that point is drop some packets so the remote end hopefully backs off a bit.
also it looked like the latency spikes co-coincided with the high CPU load instances on the RB itself, also if it was a bandwidth saturation problem I’d have expected the latency to be terrible the whole time I was downloading 
Further more the upstream was not being saturated, wasn’t even 10% loaded.