3 ms·
Given the way mine (not FritzBox) chokes due to NAT being a little stupid when burdened with enough outbound IPv4 dialing, I'd like a PCIe DOCSIS 3.1 (and enoug
by namibj 3y ago
Given the way mine (not FritzBox) chokes due to NAT being a little stupid when burdened with enough outbound IPv4 dialing, I'd like a PCIe DOCSIS 3.1 (and enough EuroDOCSIS 3.0 to get through bootstrapping, as the ISP doesn't yet support pure 3.1 moderns) modem...
I may settle for a 2.5GBASE-T "classic" modern, though, despite the lost control over the TX queue. It's sadly not possible to just presume a TX speed and forgo backpressure from DOCSIS to the packet scheduler that has the knobs to never turn Internet usage off again just to stop other more sensitive Internet usage from lagging.
- jcalvinowens 3y agoEven if you had a modem card, there are too many more potential queueing points upstream of it: what you're proposing here would never make any measurable difference in real life. The congestion that makes cable uplinks suck isn't happening at your modem. If you want to run BitTorrent and game at the same time, the router can e.g. limit the BitTorrent flows to something below the minimum guaranteed TX rate, so you know there's no queueing on the modem.
- jcalvinowens 3y agoA random anecdote: Last time I had cable was on Time Warner in the US. It was interesting because they didn't do the token bucket style ratelimiting I've seen from every ISP I've had since: they'd play games with TCP to try and slow down big flows instead. It was weird. So I wrote a little program that transferred files by sending the data over UDP at a constant rate disregarding loss, then reconciling lost bits later over TCP. That allowed me to get almost double the upload bandwidth I was theoretically paying for at the time! But on Verizon FIOS, it didn't work at all, because they really limit bandwidth at the packet level instead of trying to trick TCP into slowing down.