This is a little bit OT in my thinking, and hence I apologise upfront. This also has no direct connection with Sid's considerations for an USB cable. I never want to suggest expensive cables, for one I simply cannot afford one. This post of mine as well as the earlier one is because of my theoretical interest in the subject.
Thatguy and Ajinkya,
I am not a network professional, I am just a poor theoretical physicist. However, I have a more than casual interest in the subject (I do not want to go into the reasons here). I am delighted that you guys are professionals in the related fields and can perhaps help me understand a few things.
I know for the last decade or so that a layer-3 capable switched network with full duplex has point-to-point dedicated path and therefore collisions are avoided. Actually my question or concern is more general. The above is true when the collision domain is relatively smaller and the switching fabric is ideal. I like to know in a practical situation, like in an enterprise class LAN with 2 core switches, more than 10 distribution switches and many edge switches and several thousand switched network nodes, can collisions take place in practice, because the collision domain is now much larger and switching backplane may be saturated or some less-than-ideal network interface cards or even switches may actually send the data a bit later than what is ideal, after the carrier sensing.
Actually, routers are usually known to have something called 'late collisions'. This is pretty usual. I have personally seen that in our Institute routers, although this late collision is pretty rare but not ruled out at the core switch level too. I am sure you guys know about this. Just for reference I give a Cisco link Troubleshooting Ethernet Collisions - Cisco Systems .
There is another situation I can easily think of where at least I do not know how one can avoid collisions even in a switched network. Suppose, in my LAN, I have a MZ (militarized zone) on a private IP (say e.g. 172.X.X.X) and the MZ is connecetd to a switching device with layer-4 capabality and does NAT (Network Address Translation) and connected to the rest of the LAN through the NAT. Since the NAT converts all the IPs in my MZ to a single IP (for example), the connection from that point on cannot distinguish between packets from different nodes of the MZ. Or, can it?
A malware infected PC can send automatically billions of packets to the net and can jam the log of SQUID. This is not collision but can stall the network.
However, I like to mention that collisions are not that bad afterall, because there is usually no data loss even after there is a collision. The protocol has enough safeguard against such a thing.
I am sorry for the partly hypothetical and technical nature of this post. I have not fully understood your claims of no collisions in a Ethernet based network.
Perhaps we should take our discussion on network somewhere else or through PM, and leave Sid and his thread alone. But I find the subject highly interesting.
Ajinkya,
I fully agree that EMI may not be serious issue when it comes to a USB cable. Let alone digital, I use mostly unshielded DIY interconnects between my CDP and amp, phono-stage and amp and cassette deck and amp. The only place I make sure I use a shielded cable is between my TT and the phono, because the signals are very weak, and EMI will result in serious humming. One needs to make sure there is no significant detrimental effects due to EMI. There is certainly EMI present, and one has to be careful. We cannot ignore the subject as such, that's all. I cannot and do not buy USD 500 cables anyway.
Thatguy,
I hope you understand now that being a physicist how helpless I feel when people talk about cable burning in or the actual effect of a cable and think that this whole subject is a hoax. There are solid physics reasons behind it and also for directionality of a cable after burning in. I have tried explaining that a few places in this forum, but who cares.
Regards.
Thatguy and Ajinkya,
I am not a network professional, I am just a poor theoretical physicist. However, I have a more than casual interest in the subject (I do not want to go into the reasons here). I am delighted that you guys are professionals in the related fields and can perhaps help me understand a few things.
I know for the last decade or so that a layer-3 capable switched network with full duplex has point-to-point dedicated path and therefore collisions are avoided. Actually my question or concern is more general. The above is true when the collision domain is relatively smaller and the switching fabric is ideal. I like to know in a practical situation, like in an enterprise class LAN with 2 core switches, more than 10 distribution switches and many edge switches and several thousand switched network nodes, can collisions take place in practice, because the collision domain is now much larger and switching backplane may be saturated or some less-than-ideal network interface cards or even switches may actually send the data a bit later than what is ideal, after the carrier sensing.
Actually, routers are usually known to have something called 'late collisions'. This is pretty usual. I have personally seen that in our Institute routers, although this late collision is pretty rare but not ruled out at the core switch level too. I am sure you guys know about this. Just for reference I give a Cisco link Troubleshooting Ethernet Collisions - Cisco Systems .
There is another situation I can easily think of where at least I do not know how one can avoid collisions even in a switched network. Suppose, in my LAN, I have a MZ (militarized zone) on a private IP (say e.g. 172.X.X.X) and the MZ is connecetd to a switching device with layer-4 capabality and does NAT (Network Address Translation) and connected to the rest of the LAN through the NAT. Since the NAT converts all the IPs in my MZ to a single IP (for example), the connection from that point on cannot distinguish between packets from different nodes of the MZ. Or, can it?
A malware infected PC can send automatically billions of packets to the net and can jam the log of SQUID. This is not collision but can stall the network.
However, I like to mention that collisions are not that bad afterall, because there is usually no data loss even after there is a collision. The protocol has enough safeguard against such a thing.
I am sorry for the partly hypothetical and technical nature of this post. I have not fully understood your claims of no collisions in a Ethernet based network.
Perhaps we should take our discussion on network somewhere else or through PM, and leave Sid and his thread alone. But I find the subject highly interesting.
Ajinkya,
I fully agree that EMI may not be serious issue when it comes to a USB cable. Let alone digital, I use mostly unshielded DIY interconnects between my CDP and amp, phono-stage and amp and cassette deck and amp. The only place I make sure I use a shielded cable is between my TT and the phono, because the signals are very weak, and EMI will result in serious humming. One needs to make sure there is no significant detrimental effects due to EMI. There is certainly EMI present, and one has to be careful. We cannot ignore the subject as such, that's all. I cannot and do not buy USD 500 cables anyway.
Thatguy,
I hope you understand now that being a physicist how helpless I feel when people talk about cable burning in or the actual effect of a cable and think that this whole subject is a hoax. There are solid physics reasons behind it and also for directionality of a cable after burning in. I have tried explaining that a few places in this forum, but who cares.
Regards.
).
