Artorias it's a switch, not a router. And not a particularly good switch at that. So it's probably not relevant to compare home/SOHO switches to a commercial router. Two different types of devices and two different market segments.
never seen switches break like this. I wouldn't recommend Asus for their switches...maybe only Mobo. Get Cisco, Netgear, TP-Link for network equipment.
Had a series of that same Asus switch, all of which eventually failed one way or another. It seems to be discontinued now, probably for a good reason. But it's no big deal as decent switches are inexpensive these days.
You can disassembly and check capacitors, if they leaked you must replace them, i have same issue with netgear some ports died, and when i replace capacitors they alive and switch is working like newone :)
Here's how you fry both your network ports on a Mac Pro (or PowerMac G5): Connect your switch to your servers and Mac(s) via copper cables (anything between CAT-5 and CAT-7, I prefer CAT-6a). Hook in your neighbors on your switch as well. Make sure the phase you're powering on your Mac differs from the phases your neighbours use for their devices. The voltage difference between your neighbor and you will fry your Mac's ethernet ports. Tried it twice now. ... Solution: Make sure you run the equipment you connect via 8P8C connectors (eg. copper cables) on the same phase. When there might be a difference between phases OR when you want to really ensure that you don't want a port to fry, isolate it optically by using fiber-optics (I prefer OS2). ... Mark, I would say you're quite lucky, that you only got your switch fried. It might not be the same reason that the ports on your switch were fried, but it could be, so try checking. If you ever got an electrical shock while inserting/disconnecting copper cables, it's *very* likely that you have two (or even three) different phases on your equipment. To check if the phases are the same: If in Europe (like me), you'll likely have 3 phases; don't know about the US. 1: Set up your multimeter/voltmeter to measure 500VAC (must be at least 500V and must be AC, otherwise you could fry your multimeter/voltmeter). 2: Measure the voltage on the two prongs; one is phase (P), one is NUL (N), remember where P is and use this as the reference. 3: Measure the voltage difference on all the other plugs using the first plug's P as reference. Results: A: If the voltage is 0V, then the phase match, that is good. B: If the voltage is the same you found in step 2, then you've found a NUL, measure the other prong. C: If the voltage is higher than the voltage you found in step 2, then the phase differs. You need to find another power outlet, where the difference between the phases is 0V.
48 ports of full layer three "managed" gigabit Ethernet with redundant power supplies, etc are 30 bucks on ebay these days. Fastiron GS being pulled out of data centers for instance, even have a slot for 10gbe modules. Plenty of similar options from juniper/cisco as well but they'll be a little more pricey if you want the same ipv4/ipv6 l3 features. Ethernet is a pretty resilient spec and also 100% galvanically isolated from the ESD/discharge scenarios that typically cause sudden interface failure, if I had a single port die on any piece of networking gear I'd stop using that manufacturer altogether as that's incredibly poor design
+Jon Sands Maybe where you live they're $30, but not where I am. You are correct in that Ethernet PHYs are magnetically coupled, but that won't protect against all transients. I have seen dead ports in lots of brands like Cisco/Linksys, HP, and others - even enterprise grade equipment, so it is not **necessarily** bad design.
+Mark Furneaux Sorry, 30USD, I take it you live in canada? Most sellers will ship to canada for only 20 extra dollars or so, so it's still just whatever the current exchange rate is, 40 bucks maybe? I own an ISP in the american midwest with roughly 600 ports across several municipalities and we've never seen a physical port die, just rarely cabling (bad installers), and this is with some connected gear 150 feet up a tower, mostly brocade with some juniper at the core. I can agree with you I've seen all sorts of fun hardware failures in Cisco equipment both old and new product lines, one of them any reasons we moved away from them :)
+Jon Sands I looked quickly and the cheapest I can get is around $80, so still not *bad*, but more than what you say. That also ignores duty. I hate Cisco and haven't bought one of their products in a long time. The thing is though, I want a small desktop switch, not a huge rackmount thing. I have a rackmount switch already and it works fine.
Some ebay rack mount 16-Port or 24-Port Switches have ears that can be removed. Good Hunting... Then you have a big roomy switch with air flow and maybe a fan. tjl
Why are you using two links to the switch but only one from the switch to the core? I could see it for your file server but you said you had infiniband.
+Jay Schell You make a very good point that I should have addressed in the video. Both this switch and my core switch are unmanaged. This means that they have no support for link aggregation. If I were to connect 2 ports between the switches it would create an infinite loop between the two switches and they would ultimately disable one of the 2 links. You need managed switches to setup LAGG between switches, but I couldn't afford those when I bought the hardware. The single link is a weak point, but there isn't much I can do right now. I plan to remove the switch and just run long cables to the main switch when I get a chance. I do use Infiniband now, but when I originally built the server I used Ethernet, and never changed the wiring setup.
to aggregate the ports properly, you bond the ports in your client device, and in a smart or managed switch you set up an aggregation group. Well, thats what i do at least. You can just bridge the ports in say windows but that isn't aggregating them, teaming on intel nics is the proper term rather than bridging
In addition to Julian's reply ... It's best to get yourself a managed switch. _some_ unmanaged switches might work. Linux might be able to bond two ports via software, but this is not the same as the widely used 802.3ad (preferred) aggregation supported by most equipment. I use D-Link DGS-1510-20 (any D-Link managed switch will do fine; I don't know about other brands). Some switches/routers might be able to get firmware upgrades (including DD-WRT/OpenWRT), but ... their built-in switches might not be able to give you more than 1Gbit/sec in total, because the interface between the switch and the router's CPU might be only 1Gbit. Some devices (servers/routers - not talking switches here) are difficult to set up, even if you're using Linux - this is due to the way their hardware is connected on the PCB. In particular, I'm thinking about devices where the switch is connected via a 1Gbit connection to the CPU.
Picked up a Dell 24-Port Switch in a 1U Case from ebay. And up-graded to some Sunnto quiet fans. Have you replaced your Switch ? How is your Web Modem holding up ? Thank you for the video ! tjl
I use balance-rr on two servers for quite a while by now, never noticed any negative performance effects. One direction iperf is 920 Mbit/sec, both directions at the same time is 850+850 Mbit. I'd say this "out of order" issue is way overblown, at least if you just connect both ports into the same switch, and not doing something more complex (such as going over leased long distance links, or even over the Internet).
hey, have you did the port trunking on the switch itself and on the workstation? I have a qnap, port trunking it with IEEE 802.3ad, both knows this protocol, but from 1Gb internet speed I only get :) 48Mb ... Haven`t change the ports in the switch yet, dont seems to be dead or something.
What do you do for a living? Also i had some computer Ideas i want someone with more cash than i have to test like using a raspberry to control a fan controller board with a system monitor to auto turn things up when i used more resources. shit like that.
Heh, amusing that it took over a year for you to do a video on how shit Marvell is! We have around 40 or 50 machines running Debian provisioned on some OEM mini desktop, all running fucking Marvell based NIC's. It was within the space of 3 or 4 months that at least half either died completely, or oddly enough, I'd have to bring the iface down and up. I would have a watchdog script to ifdown eth0 ifup eth0 when it can't ping the router. In short, fuck Marvell.
I would get a new switch and make it a managed switch. They are fairly inexpensive for an 8-port D-Link is about $100. I WENT FOR A 24 PORT Managed SMC for $250. When it comes to hardware you get what you pay for. My next upgrade is 24-port 10Gb per port switch from Supermicro. Its all Cat 6 only network cable connections and it is about $12K. But I should be good for the rest of my life with it. I should have it by the end of the year.
+BjergTv All my machines run Linux. Windows does not natively support LAGG, so the only way to do it is if the driver for your network card comes with the feature built-in. Only enterprise grade server and workstation NICs seem to have support.
This is reason why I use commercial routers instead of home routers.
Artorias it's a switch, not a router. And not a particularly good switch at that. So it's probably not relevant to compare home/SOHO switches to a commercial router. Two different types of devices and two different market segments.
never seen switches break like this. I wouldn't recommend Asus for their switches...maybe only Mobo. Get Cisco, Netgear, TP-Link for network equipment.
Had a series of that same Asus switch, all of which eventually failed one way or another. It seems to be discontinued now, probably for a good reason. But it's no big deal as decent switches are inexpensive these days.
You can disassembly and check capacitors, if they leaked you must replace them, i have same issue with netgear some ports died, and when i replace capacitors they alive and switch is working like newone :)
That's cool. It doesn't bottle neck at the ISP? What's the combined bitrate? Would 1 cat6a not suffice? Just curious
Here's how you fry both your network ports on a Mac Pro (or PowerMac G5):
Connect your switch to your servers and Mac(s) via copper cables (anything between CAT-5 and CAT-7, I prefer CAT-6a).
Hook in your neighbors on your switch as well. Make sure the phase you're powering on your Mac differs from the phases your neighbours use for their devices.
The voltage difference between your neighbor and you will fry your Mac's ethernet ports. Tried it twice now.
... Solution: Make sure you run the equipment you connect via 8P8C connectors (eg. copper cables) on the same phase.
When there might be a difference between phases OR when you want to really ensure that you don't want a port to fry, isolate it optically by using fiber-optics (I prefer OS2).
... Mark, I would say you're quite lucky, that you only got your switch fried.
It might not be the same reason that the ports on your switch were fried, but it could be, so try checking.
If you ever got an electrical shock while inserting/disconnecting copper cables, it's *very* likely that you have two (or even three) different phases on your equipment.
To check if the phases are the same:
If in Europe (like me), you'll likely have 3 phases; don't know about the US.
1: Set up your multimeter/voltmeter to measure 500VAC (must be at least 500V and must be AC, otherwise you could fry your multimeter/voltmeter).
2: Measure the voltage on the two prongs; one is phase (P), one is NUL (N), remember where P is and use this as the reference.
3: Measure the voltage difference on all the other plugs using the first plug's P as reference.
Results:
A: If the voltage is 0V, then the phase match, that is good.
B: If the voltage is the same you found in step 2, then you've found a NUL, measure the other prong.
C: If the voltage is higher than the voltage you found in step 2, then the phase differs. You need to find another power outlet, where the difference between the phases is 0V.
48 ports of full layer three "managed" gigabit Ethernet with redundant power supplies, etc are 30 bucks on ebay these days. Fastiron GS being pulled out of data centers for instance, even have a slot for 10gbe modules. Plenty of similar options from juniper/cisco as well but they'll be a little more pricey if you want the same ipv4/ipv6 l3 features. Ethernet is a pretty resilient spec and also 100% galvanically isolated from the ESD/discharge scenarios that typically cause sudden interface failure, if I had a single port die on any piece of networking gear I'd stop using that manufacturer altogether as that's incredibly poor design
+Jon Sands Maybe where you live they're $30, but not where I am. You are correct in that Ethernet PHYs are magnetically coupled, but that won't protect against all transients. I have seen dead ports in lots of brands like Cisco/Linksys, HP, and others - even enterprise grade equipment, so it is not **necessarily** bad design.
+Mark Furneaux Sorry, 30USD, I take it you live in canada? Most sellers will ship to canada for only 20 extra dollars or so, so it's still just whatever the current exchange rate is, 40 bucks maybe? I own an ISP in the american midwest with roughly 600 ports across several municipalities and we've never seen a physical port die, just rarely cabling (bad installers), and this is with some connected gear 150 feet up a tower, mostly brocade with some juniper at the core. I can agree with you I've seen all sorts of fun hardware failures in Cisco equipment both old and new product lines, one of them any reasons we moved away from them :)
+Jon Sands I looked quickly and the cheapest I can get is around $80, so still not *bad*, but more than what you say. That also ignores duty. I hate Cisco and haven't bought one of their products in a long time. The thing is though, I want a small desktop switch, not a huge rackmount thing. I have a rackmount switch already and it works fine.
Some ebay rack mount 16-Port or 24-Port Switches have ears that can be removed. Good Hunting... Then you have a big roomy switch with air flow and maybe a fan. tjl
I noticed my 3G modem takes long time boot up
6 minutes and no explanation of why to use it.
Exacly
@@AFiB1999 This was the feel I got @12 seconds in and paused. Thank you for letting me know not to waste the rest of my time.
I think I have the same switch as you. Does lagg perform well on it?
Why are you using two links to the switch but only one from the switch to the core? I could see it for your file server but you said you had infiniband.
+Jay Schell You make a very good point that I should have addressed in the video. Both this switch and my core switch are unmanaged. This means that they have no support for link aggregation. If I were to connect 2 ports between the switches it would create an infinite loop between the two switches and they would ultimately disable one of the 2 links. You need managed switches to setup LAGG between switches, but I couldn't afford those when I bought the hardware. The single link is a weak point, but there isn't much I can do right now. I plan to remove the switch and just run long cables to the main switch when I get a chance. I do use Infiniband now, but when I originally built the server I used Ethernet, and never changed the wiring setup.
The dead ports, were link LAG enabled on them when they died?
How do you set up link aggregation? Is a switch required?
to aggregate the ports properly, you bond the ports in your client device, and in a smart or managed switch you set up an aggregation group. Well, thats what i do at least. You can just bridge the ports in say windows but that isn't aggregating them, teaming on intel nics is the proper term rather than bridging
In addition to Julian's reply ...
It's best to get yourself a managed switch.
_some_ unmanaged switches might work.
Linux might be able to bond two ports via software, but this is not the same as the widely used 802.3ad (preferred) aggregation supported by most equipment.
I use D-Link DGS-1510-20 (any D-Link managed switch will do fine; I don't know about other brands).
Some switches/routers might be able to get firmware upgrades (including DD-WRT/OpenWRT), but ... their built-in switches might not be able to give you more than 1Gbit/sec in total, because the interface between the switch and the router's CPU might be only 1Gbit.
Some devices (servers/routers - not talking switches here) are difficult to set up, even if you're using Linux - this is due to the way their hardware is connected on the PCB. In particular, I'm thinking about devices where the switch is connected via a 1Gbit connection to the CPU.
Picked up a Dell 24-Port Switch in a 1U Case from ebay. And up-graded to some Sunnto quiet fans. Have you replaced your Switch ? How is your Web Modem holding up ? Thank you for the video ! tjl
Strange question, in what way are these ports just "dying" is it a mechnical thing, or is it something internal?
It's purely electrical. It's probably the port driver transceivers burning out.
what's the model on that router? looks like my old WL-520gU
. oh and what causes the ports to go bad? i've never had that happen w/ any of my routers.
Doesn't round-robin cause out-of-order packets?
Yes it does, but TCP takes care of that.
Doesn't it impact performance though?
Not enough for me to care. It only affects latency. Bandwidth will still be higher.
I use balance-rr on two servers for quite a while by now, never noticed any negative performance effects. One direction iperf is 920 Mbit/sec, both directions at the same time is 850+850 Mbit. I'd say this "out of order" issue is way overblown, at least if you just connect both ports into the same switch, and not doing something more complex (such as going over leased long distance links, or even over the Internet).
That is what tcp's function is it makes sure the packets get there and other layers of the osi model handles issues like that
hey, have you did the port trunking on the switch itself and on the workstation?
I have a qnap, port trunking it with IEEE 802.3ad, both knows this protocol, but from 1Gb internet speed I only get :) 48Mb ...
Haven`t change the ports in the switch yet, dont seems to be dead or something.
What do you do for a living? Also i had some computer Ideas i want someone with more cash than i have to test like using a raspberry to control a fan controller board with a system monitor to auto turn things up when i used more resources. shit like that.
Heh, amusing that it took over a year for you to do a video on how shit Marvell is! We have around 40 or 50 machines running Debian provisioned on some OEM mini desktop, all running fucking Marvell based NIC's. It was within the space of 3 or 4 months that at least half either died completely, or oddly enough, I'd have to bring the iface down and up. I would have a watchdog script to ifdown eth0 ifup eth0 when it can't ping the router. In short, fuck Marvell.
I would get a new switch and make it a managed switch. They are fairly inexpensive for an 8-port D-Link is about $100. I WENT FOR A 24 PORT Managed SMC for $250. When it comes to hardware you get what you pay for. My next upgrade is 24-port 10Gb per port switch from Supermicro. Its all Cat 6 only network cable connections and it is about $12K. But I should be good for the rest of my life with it. I should have it by the end of the year.
i have a question what os are you using because i cannot get link aggregation on windows 10 pro...
+BjergTv All my machines run Linux. Windows does not natively support LAGG, so the only way to do it is if the driver for your network card comes with the feature built-in. Only enterprise grade server and workstation NICs seem to have support.
Mark Furneaux Awesome That explains ill have a look at linux and maybe run windows with qemu ty for your answer! and nice video :D
+Mark Furneaux OS X has native LAG support
+BjergTv can you not use network teaming? /watch?v=CfUjy3Bvqvw
+Rick Deckard does not work in windows 10 anymore after and update microshaft winblows made so i had to revert to windows 8.1......