On Mon, 8 Feb 2021 16:42:40 -0500 Dan Ritter <d...@randomstring.org> wrote:
> Celejar wrote: > > > If you are OK buying used equipment, Intel-based gigabit NICs, 4 ports > > > to a PCIe slot, cost about $35 (or $70 new). If you've got a 5 year old > > > > My understanding - please correct me if I'm wrong - is that with those > > types of cards, the ports are distinct and aren't actually switched in > > hardware, so switching occurrs at the OS / kernel level. I don't know > > how much of a load this puts on the system in practice, but my > > understanding is that it's certainly not an ideal way to design a > > switch. > > Modern processors -- even the ones 5 years old -- are really > fast. > > Linux bridging (switching) is very efficient. Fair enough. > Is it "ideal"? No. But given that you want one device which acts > as a WAP, router, firewall and switch, it should perform quite > well. If you hate the idea of doing that, though, an 8-port > gigabit switch is about the same price as a used 4-port gigabit > NIC. Not as flexible, though. > > > > desktop sitting around with 2GB or more RAM and 3 available PCIe slots, > > > you can use it as a WAP and have nine switched/routed gigabit ports, > > > counting one on the motherboard. If you only need 5 ports, you only > > > need 2 PCIe slots -- one for a WiFI NIC and one for the ethernet NIC. > > > > My understanding, although I could not find solid documentation of this, > > is that consumer wireless chipsets designed for client use don't make > > particularly performant APs. They'll work, but purpose built APs will > > perform much better, especially with their AP optimized antennas. I > > don't really know if this is true, though, and to what extent it's an > > issue, if it really is one. > > Oh, no, this is a myth. The $20-150 consumer wifi routers use > the same wifi interface chips as good PCIe cards, for the most > part. OpenWRT is actually a great source of information on > these. > > Assuming you're comparing a 3 antenna MIMO on a PCIe card to a 3 > antenna MIMO on a consumer router, you should get equivalent > range and performance. Thanks. I'd love to see actual tests comparing performance of wireless APs (consumer, enterprise, and DIY ones like we're discussing), but they seem very hard to come by. > > And the power usage on a five year old desktop (which I don't actually > > have) will be much higher than a purpose-built AIO AP / switch / router. > > That can be true. But then, the desktop can also be your server > for a bunch of other things that, perhaps, you were going to > run. Fair enough. I'm currently using an old R210 ii as my server, so I'm not one to talk ;) I suppose it might be fun to see if I can fit a modern AX200 based PCIe (perhaps a low profile one) into it and see how it performs as an AP / router ... > > But again, I don't really disagree. If I had the hardware lying around, > > and I determined that the power consumption wasn't a factor, it would > > certainly be tempting to consider this route. > > Everything is a tradeoff. Yes. Celejar
