TR Forums

Waco wrote:

You only have to route if you're leaving the subnet (going to the Internet) - dumb switches route between machines on the same subnet without having to hit the router.

I knew there was a simple answer my aging brain just hadn't figured out. Like I said, I don't grok the OSI model, so I was having problems figuring out why a device without an IP address could accomplish this.

I know enough about networking to look at a diagram (in the day job) and ask "please explain that decision" (us regulators love open-ended questions). Isn't a whelk's chance in a supernova I'll ever get any networking certs.

I wouldn't hold my breath. Streaming, cloud, Wi-Fi and mobile has made wired LAN more or less irrelevant to the average user, and I dare say the vast majority don't have broadband plans that saturates a 1gbps WAN port. Whatever demand for 2.5/5Gbps is, its gonna be tepid at best.

Captain Ned wrote:

Waco wrote:

You only have to route if you're leaving the subnet (going to the Internet) - dumb switches route between machines on the same subnet without having to hit the router.

I knew there was a simple answer my aging brain just hadn't figured out. Like I said, I don't grok the OSI model, so I was having problems figuring out why a device without an IP address could accomplish this.

I know enough about networking to look at a diagram (in the day job) and ask "please explain that decision" (us regulators love open-ended questions). Isn't a whelk's chance in a supernova I'll ever get any networking certs.

I'm sorry but that's not accurate. "Dumb" switches do switching not routing. If the destination MAC is in their address table they send the packet to those port(s). They will never see, nor care about IP addresses.

Now the subnet still matters because if the computer doesn't have a direct route to the device it's destination will be your default gateway (router), but there are various tricks you could do to avoid hitting the slow router interface like putting one of the hosts in two subnets.

BobbinThreadbare wrote:

Captain Ned wrote:

Waco wrote:

You only have to route if you're leaving the subnet (going to the Internet) - dumb switches route between machines on the same subnet without having to hit the router.

I knew there was a simple answer my aging brain just hadn't figured out. Like I said, I don't grok the OSI model, so I was having problems figuring out why a device without an IP address could accomplish this.

I know enough about networking to look at a diagram (in the day job) and ask "please explain that decision" (us regulators love open-ended questions). Isn't a whelk's chance in a supernova I'll ever get any networking certs.

I'm sorry but that's not accurate. "Dumb" switches do switching not routing. If the destination MAC is in their address table they send the packet to those port(s). They will never see, nor care about IP addresses.

Now the subnet still matters because if the computer doesn't have a direct route to the device it's destination will be your default gateway (router), but there are various tricks you could do to avoid hitting the slow router interface like putting one of the hosts in two subnets.

Sorry for not using the exactly correct word (I should have said pass instead of route). I understand how MACs work, but not many consumers really care how switching/routing is done, only what the effects are locally and how it affects them.

For 99.99999% of consumer home networks, there's a subnet for local and then there's everything else - so I was attempting, poorly, to put it into layman's terms.

Waco wrote:

Sorry for not using the exactly correct word (I should have said pass instead of route). I understand how MACs work, but not many consumers really care how switching/routing is done, only what the effects are locally and how it affects them.

For 99.99999% of consumer home networks, there's a subnet for local and then there's everything else - so I was attempting, poorly, to put it into layman's terms.

Global addressing in IPv6 is going to be *sheer fun* for people with home networks that don't understand the details. Internet of (hackable) Things, indeed. I wonder how many severe consumer firewall bugs are masked by PAT? I guess we're going to find out!

Yeah, IPv6 is a whole new wrench in the works.

Waco wrote:

Yeah, IPv6 is a whole new wrench in the works.

Well, with 1.5 * 10^18 IPv6 addresses for every square millimeter of the Earth's surface (oceans included), what could possibly go wrong?

Vhalidictes wrote:

Waco wrote:

Sorry for not using the exactly correct word (I should have said pass instead of route). I understand how MACs work, but not many consumers really care how switching/routing is done, only what the effects are locally and how it affects them.

For 99.99999% of consumer home networks, there's a subnet for local and then there's everything else - so I was attempting, poorly, to put it into layman's terms.

Global addressing in IPv6 is going to be *sheer fun* for people with home networks that don't understand the details. Internet of (hackable) Things, indeed. I wonder how many severe consumer firewall bugs are masked by PAT? I guess we're going to find out!

I suspect most people will probably still be behind a NAT. You *can* do universal addressing, but why not just have the router have the IPv6 address and NAT it to actually readable IPv4 addresses.

Although one cool thing you could do would be selective pass through of IPv6 which would help people with multiple Xboxes/PS4s etc. Like a DMZ but as many as you want.

I'd shove a 2-port 10Gb card in a PC with a high-mhz CPU and run pfSense on it before trying the early adopter lottery on consumer routers.

BobbinThreadbare wrote:

Vhalidictes wrote:

Waco wrote:

Sorry for not using the exactly correct word (I should have said pass instead of route). I understand how MACs work, but not many consumers really care how switching/routing is done, only what the effects are locally and how it affects them.

For 99.99999% of consumer home networks, there's a subnet for local and then there's everything else - so I was attempting, poorly, to put it into layman's terms.

Global addressing in IPv6 is going to be *sheer fun* for people with home networks that don't understand the details. Internet of (hackable) Things, indeed. I wonder how many severe consumer firewall bugs are masked by PAT? I guess we're going to find out!

I suspect most people will probably still be behind a NAT. You *can* do universal addressing, but why not just have the router have the IPv6 address and NAT it to actually readable IPv4 addresses.

Although one cool thing you could do would be selective pass through of IPv6 which would help people with multiple Xboxes/PS4s etc. Like a DMZ but as many as you want.

You just setup a simple firewall rule that says connections must be initiated from the LAN port, only related packets are allowed back in on the WAN port. No need for DMZ or anything silly like that.

BobbinThreadbare wrote:

I suspect most people will probably still be behind a NAT. You *can* do universal addressing, but why not just have the router have the IPv6 address and NAT it to actually readable IPv4 addresses.

Although one cool thing you could do would be selective pass through of IPv6 which would help people with multiple Xboxes/PS4s etc. Like a DMZ but as many as you want.

Yeah, there are tons of ways to handle IPv6 connections. There's NAT/PAT support built-in, but it's really up to how the ISP handles it. From what I've seen so far they are giving out semi-private (globally addressable but are assigned to them) IP ranges in their space. But Comcast has been giving out fully non-route-able addresses... so far...

notfred wrote:

You just setup a simple firewall rule that says connections must be initiated from the LAN port, only related packets are allowed back in on the WAN port. No need for DMZ or anything silly like that.

I guess that modern OSs are secure enough that this might work for a self-installed FW. Of course, then again, who knows what D-link et al will end up doing in their security code.

SuperSpy wrote:

I'd shove a 2-port 10Gb card in a PC with a high-mhz CPU and run pfSense on it before trying the early adopter lottery on consumer routers.

We're a long way from needing 10 Gb routing, though.

notfred wrote:

You just setup a simple firewall rule that says connections must be initiated from the LAN port, only related packets are allowed back in on the WAN port. No need for DMZ or anything silly like that.

There are a lot of customer applications that require the ability to open connections from WAN (thus the invention of UPNP). Modern game consoles are the most common since they use p2p instead of dedicated servers.

People run into problems with multiple devices trying to use the same ports which is why the ability to pass through IPv6 addresses at will would be cool.

Bauxite wrote:

Dead in the water until prices are not idiotic.

It is literally cheaper to build a 40Gbit network off ebay (switches, optics, fiber and all) than to upgrade your network to 10Gbe copper and hope it works over your short cat6 runs. It might even use less power

This. And if you want 10G, you can get that with new equipment if you just go with fiber. It's also cheaper. For instance the Ubiquiti ES-16-XG Edge Switch gives you 12 SFP+ ports that can handle 10G speeds (and 4 RJ45 if you really need them for something), and it is $511 at this time. It also is a managed layer 3 switch. Far superior to to that Buffalo offering. Getting 10G fiber network cards can be had for about the same price new as RJ45 ones, or you can get used ones for cheaper.

If you want to go with EBAY, like Bauxite suggests, you will probably spend a couple hundred more than the Buffalo option, but you get 4 times the bandwidth. I haven't looked into the SFP cost though (for 10G with Ubiquiti at least, the SFPs are cheap).

10G RJ45 just isn't ready yet, imho. It needs a couple more years.

notfred wrote:

BobbinThreadbare wrote:

Vhalidictes wrote:

Global addressing in IPv6 is going to be *sheer fun* for people with home networks that don't understand the details. Internet of (hackable) Things, indeed. I wonder how many severe consumer firewall bugs are masked by PAT? I guess we're going to find out!

I suspect most people will probably still be behind a NAT. You *can* do universal addressing, but why not just have the router have the IPv6 address and NAT it to actually readable IPv4 addresses.

Although one cool thing you could do would be selective pass through of IPv6 which would help people with multiple Xboxes/PS4s etc. Like a DMZ but as many as you want.

You just setup a simple firewall rule that says connections must be initiated from the LAN port, only related packets are allowed back in on the WAN port. No need for DMZ or anything silly like that.

You ought to have a DMZ though if you are setting up any services that can be initiated from outside your home network, which, given IoT devices, is becoming a pretty big deal even for regular consumers. Unfortunately, it is not easy to set one up properly.

still nada... very disappointing!

Dunno about routers, but the ASUS XG-C100C that launched is a ~$100 10Gbps NIC. It also specifically supports 2.5GBASE-T and 5GBASE-T links.

Given ASUS already makes some 10GBASE-T switches I would tend to think that means ASUS will be launching some updated consumer routers within the year that people can pair with their own 10G NIC.

Kougar wrote:

Dunno about routers, but the ASUS XG-C100C that launched is a ~$100 10Gbps NIC. It also specifically supports 2.5GBASE-T and 5GBASE-T links.

Given ASUS already makes some 10GBASE-T switches I would tend to think that means ASUS will be launching some updated consumer routers within the year that people can pair with their own 10G NIC.

Yes, but if they are going to cost ~$80 per 10GBASE-T port, then they are more expensive than fiber solutions where you could get a managed switch with layer 3 capability for the same price and it will have more ports to boot.

I think we'll have to wait 2 years or so (at a guess) before the price gets to a reasonable level and becomes much cheaper than fiber.

Edit: I don't mean to harp on this overmuch, but unless you have extensive CAT6 cabling already ran and it would be hard to run fiber, then it is something everyone should be aware of.

Drachasor wrote:

Yes, but if they are going to cost ~$80 per 10GBASE-T port, then they are more expensive than fiber solutions where you could get a managed switch with layer 3 capability for the same price and it will have more ports to boot.

They shouldn't need to, the cost of the internals wouldn't be that much per port. The guts of a L3 switch are enough to make a good router already, there's hardly a difference anymore beyond the label. A lot of server farms have already replaced all their older internal routers with L3 switches because there's no longer a difference except for price.

Consumer routers are actually L3 switches anyway. Case in point, a server router connected directly to a client requires a crossover cable. But we consumers don't buy crossover cables to connect our desktops to our home "routers". Granted modern server networking gear can sense and swap pins 1+2 with 3+6 so either cable can be used (no idea how common it is for home "routers"), but I remember this precisely because I got this question wrong on a practice exam and had to bug the professor about it.

Drachasor wrote:

Edit: I don't mean to harp on this overmuch, but unless you have extensive CAT6 cabling already ran and it would be hard to run fiber, then it is something everyone should be aware of.

The idea is the 2.5 and 5GBase-T standard allows existing Cat5e to automatically upgrade to 2.5Gbps over 100m. In fact, if you use less than 50m you can expect to attain 5GBASE-T over Cat5e cabling. That is absolutely perfect for existing home networks (I'm not talking home installations, just behind the desk).

All I want is 5G from my system to my NAS for my weekly SSD-image backups. I run daily incremental backups, but SSD images take awhile to transfer even though most M.2 SSDs could still max out a 5GBASE-T connection.

Kougar wrote:

Drachasor wrote:

Yes, but if they are going to cost ~$80 per 10GBASE-T port, then they are more expensive than fiber solutions where you could get a managed switch with layer 3 capability for the same price and it will have more ports to boot.

They shouldn't need to, the cost of the internals wouldn't be that much per port. The guts of a L3 switch are enough to make a good router already, there's hardly a difference anymore beyond the label. A lot of server farms have already replaced all their older internal routers with L3 switches because there's no longer a difference except for price.

It was just an estimate based on the one Buffalo switch. The point remains right now it is ridiculously expensive to use 2.5/5/10BASE-T, and you might as well go fiber instead. Or wait a few years.
Heck, if you want more bandwidth for a couple devices, then it IS much cheaper to just aggregate some connections. 4+ 1G ports can be done with a managed switch on each end, and you can do all that AND ethernet for less than the Buffalo switch. Hmm, if you want really good managed 1G switches, that's about $200 each (you can do it for half that and get decent stuff or good used stuff), and maybe $50 for each 4-port card. That's $500 total if new gear, and more like $300 if used. Now, granted, you will need to then have an aggregate connection between the switches that also uses 4 ports on each. Unless you got switches that have SFPs that support 10G or the like on 2-4 ports for connecting to other network gear. Then you need the SFPs, but those can be pretty cheap too.

All in all, I can't see a good argument for getting into 2.5-10BASE-T right now. There are far more affordable alternatives however you slice it.

Kougar wrote:

Dunno about routers, but the ASUS XG-C100C that launched is a ~$100 10Gbps NIC. It also specifically supports 2.5GBASE-T and 5GBASE-T links.

Given ASUS already makes some 10GBASE-T switches I would tend to think that means ASUS will be launching some updated consumer routers within the year that people can pair with their own 10G NIC.

a switch is all well and good, but when the center of my connections is still stuck at 1G, isn't not that helpful

will be a good pickup when someone finally releases one

leor wrote:

Kougar wrote:

Dunno about routers, but the ASUS XG-C100C that launched is a ~$100 10Gbps NIC. It also specifically supports 2.5GBASE-T and 5GBASE-T links.

Given ASUS already makes some 10GBASE-T switches I would tend to think that means ASUS will be launching some updated consumer routers within the year that people can pair with their own 10G NIC.

a switch is all well and good, but when the center of my connections is still stuck at 1G, isn't not that helpful

will be a good pickup when someone finally releases one

Why does the router interface matter at all? If you don't have more than 1 Gbps of Internet connectivity, it's literally a non-issue.

Waco wrote:

leor wrote:

Kougar wrote:

Dunno about routers, but the ASUS XG-C100C that launched is a ~$100 10Gbps NIC. It also specifically supports 2.5GBASE-T and 5GBASE-T links.

Given ASUS already makes some 10GBASE-T switches I would tend to think that means ASUS will be launching some updated consumer routers within the year that people can pair with their own 10G NIC.

a switch is all well and good, but when the center of my connections is still stuck at 1G, isn't not that helpful

will be a good pickup when someone finally releases one

Why does the router interface matter at all? If you don't have more than 1 Gbps of Internet connectivity, it's literally a non-issue.

The router assigns the internal IP addresses, no? Doesn't all internal traffic pass through the router at some point? Wouldn't the slowest link in the chain constrict all traffic to that speed?

leor wrote:

The router assigns the internal IP addresses, no? Doesn't all internal traffic pass through the router at some point? Wouldn't the slowest link in the chain constrict all traffic to that speed?

Yes, it would. But if your Internet pipe is about 1Gbit, it doesn't really matter how much higher your throughput is... on the Internet-facing port.

Most network devices have sufficient throughput for their ports, so I'd assume that 10GBASE-T ports could talk to other 10GBASE-T ports at wire-speed. It really depends on how the internal design of the router backplane.

Yes, the DHCP server in your router is usually what assigns your local IP addresses, but that doesn't mean all internal traffic is required to go through the router. Traffic between clients on your local network can often go through only the switch and bypass the router completely.

leor wrote:

The router assigns the internal IP addresses, no? Doesn't all internal traffic pass through the router at some point? Wouldn't the slowest link in the chain constrict all traffic to that speed?

Only internal traffic that needs to leave your local network is routed through your routers WAN port. Switches can run without any router at all for just local access.

Kougar wrote:

But we consumers don't buy crossover cables to connect our desktops to our home "routers". Granted modern server networking gear can sense and swap pins 1+2 with 3+6 so either cable can be used (no idea how common it is for home "routers"), but I remember this precisely because I got this question wrong on a practice exam and had to bug the professor about it.

That's actually a requirement of the gigabit PHY spec so it the problem solves itself on modern networking hardware.

SuperSpy wrote:

That's actually a requirement of the gigabit PHY spec so it the problem solves itself on modern networking hardware.

Actually it isn't (it's optional according to the spec), but I've never seen hardware without it.

Autonegiation *is* required, but that only deals with speed/duplex, not MDI/MDI-X.

Waco wrote:

leor wrote:

The router assigns the internal IP addresses, no? Doesn't all internal traffic pass through the router at some point? Wouldn't the slowest link in the chain constrict all traffic to that speed?

Only internal traffic that needs to leave your local network is routed through your routers WAN port. Switches can run without any router at all for just local access.

gotcha, so, not like the old SCSI interface

so connect the router to the switch and connect all the devices with fast network connections to the switch only and I'm golden

still would be nice if this tech started making its way into routers, the standard has been out for damn near a year!