IOS upgradation of switch, router and pix

snipped-for-privacy@gmail.com wrote:

Can someone explain (in one line!) the difference between a Collision domain and a Broadcast domain

Collision domain is the group of hosts in which collision can occur, and broadcast domain consists of all the groups of hosts that can proceed the broadcast frame.

Doan

It's not that simple as on does not necessarily exclude the other. A collision domain by its very nature is also a broadcast domain but a broadcast domain may or may not be a collision domain.

A collision domain is one in which multiple hosts can potentially transmit at the same time like multiple hosts connected to a hub (half-duplex Ethernet) . A broadcast domain is one in which all hosts will see 'broadcast' frames like multiple hosts connected to a switch whose ports are in the same vlan. A device that performs layer 3 switching typically separates broadcast domains.

BernieM

Hub vs. router.

Jonathan

Collisions happen at layer 1, within CSMA/CD media. Broadcasts happen at layer

1. A collision domain is contained entirely within a contiguous group of layer-1 devices. (10baseT hubs, coax, transcievers) When you reach a device that operates at layer 2 or above, that is the edge of the collision domain. A broadcast domain is contained within layer-1 and layer-2 devices (hubs, switches, bridges), and is bounded by anything that operates at layer 3 or above (router, firewall, content switch, etc.)

OK, that was more than one line, but understanding technology is not about one-liners.

A VLAN is a single broadcast domain but some of the hosts in that VLAN might also be in a collision domain. If we're talking about a switch for example, any host that has a half-duplex connection to the switch, if they're in the same VLAN they're also in the same collision domain.

Some might disagree and say each of those hosts has their own unique collision domain ... the switch itself being the only other member but I'm considering the fact that the switch is simple forwarding another hosts packet and if that's out another half-duplex port then it's a host-to-host collision not a host-to-switch collision that's taken place.

BernieM

"The Switch Book" by Seifert says that a layer 3 switch is a router. Just cisco market speak. Of course cisco would like the term 'layer 3 switch' used. And switch is market-speak for bridge. Cisco distinguish saying switches use ASICs, but that's just cisco. Really, functionally, it is what it is.

I'm still very much a newbie, studying for CCNA (though I do read), and, I ran into a problem. Why are the border addresses of the old Classful addressing system , 'Reserved'. RFCs weren't so helpful in telling me, but 2 posts on usenet helped. Yours and a guy whose surname was saunders. (The others gave wrong inconsistent reasons based on a poor failed attempt at a psychological explanation!).

I got that Routers don't like it if after they read the class identification bits they see All 0s and All 1s. That explains why those addresses are hit. But why don't they like it?

Clearly, this is only a problem with Classful addressing?

and,

[here i'll include some of my thinking, I hope it doesn't get too waffly, all my thinking fails to reach an answer]

I dont think it's to do with Network 0 being "this network". Because the network number / network ID includes the class identification bits. The Router should Read the Class identification bits but not strip them off. Since a packet from a Class C network settting teh network number to 0 would refer to "this network" it wouldn't set the network number to 192. Those Class ID bits are part of the network number.

I notice Class A is slightly different, it reserves the first and last networks for those well known special meanings. Whereas the old Classes B and C border addresses are reserved for that unspecified reason under discussion here, whuich i'm trying to figure out.

The old Class D and E border addresses are reserved( though I notice that RFC 3330 was written in 2002 and uses CIDR notation - but oddly says /4 rather than /32).

I know that now they're only reserved for historical reasons and will soon or have been assigned or are subject for assignment. So, my main q is why - in the classful days - was it an issue?

Many thanks for your posts, I've learnt a lot from them.

I disagree. You are confusing the operation of a hub and that of a switch. A hub is just a pass-through device and thus you have this big collision domain. A switch, however does micro-segmenation; each port is a collision domain, in half-duplex. In a collision domain, only one host can talk at a time. A switch, however, supports many conversations simultanously.

Doan

Technically- and in an old fashioned sense the term hub doesn't just mean Repeater or multiport repeater. A repeater is only one form of hub. It comes from 'hub and spoke topology', so, in theory, the term hub can apply any network device that is central with > 1 socket, where other devices connect to it with cables. Like a star shape. Indeed, I think a star topology is another name for 'hub and spoke topology'. In that sense, a router is a hub. But nobody says that nowadays. Before the 802.3 LAN Technology made it big, many people had Token Ring 802.5. With that, Computers were attached with cables to a central device called an MAU, this was referred to by many, as a hub. Even though it was not a repeater.

However, that is v. old fashioned terminology. in reality. Nowadays, A Hub has come to only mean either a 'passive hub' or an 'active hub'. An active hub boosts the signal. A passive hub, does not. Either way, they have > 1 port, they receive the signal at one port, and send it out of all the other ports, but not back out the one it came in. More correctly, nowadays, a hub only means a repeater. A post by Richard Seifert, to my recollection, quoted the 802 spec which says that - to paraphrase - a repeater is either a device that boosts the signal or adds connectivity to a network.

So nowadays, a hub is a repeater. And a repeater receives data in one port and sends it out all the others but not the one the data came in on.

I was told by a CCIE that the hub actually has a chip at each port to prevent the signal from coming back out from where it came. So, the hub isn't just a shared bus inside. I haven't got a book/textual source for that though. All my books claim it's a shared bus inside, and then say that the signal doesn't come out the port it came in.

What is a Router? Well, first you should really know what a Bridge(or switch) is.

I could explain - and will explain somewhat, but, have you looked at any books? Do you know what a MAC Address is?

When I was 16, I used to sit back in computer classes and basically say "teach" and I was a star pupil simply for listening and understanding, the teacher was just glad somebody was listening. Now, once you get in the real world, if say, you try to learn unix, and you sit aroudn and say "teach me" they'll say RTFM!! You really should indicate that you have read stuff and at least that you have a problem with something you've read!

I'll tell you, the basic difference between a hub and a switch, is that if computers A,B,C,D and E are connected to a Hub, and Computer A transmits data(a signal) intending it to go to D, then it goes to the hub, and then to B,C and D. D will accept it. B and C will reject it. Not very efficient, is it!! Ideally, you want A to E direct. If Comp A sends data(a frame actually) intended for E, then the switch will forward it from A to E without sending it to all others - unless A tells it to.

I could be Very wrong in a few parts here, if so I apologise and hope I am corrected. I am studying for a CCNA, my knowledge is small but growing slowly. I'm relatively rigorous. All theory and no practice - yet.

q_q_

-

I understand how a switch and hub operate and am not confused by the operation of either. What I was implying was the 'grey area' between half-duplex ports on a switch in regard to them being within the same collision domain or not. If you look at the collision itself, the mac addresses of the colliding frames identify unique hosts ... each on different switch ports. So technically speaking haven't 'their' transmissions collided. How could that occur unless they were in the same collision domain.

I understand the three main caterories of switch operation ie. 'store-and-foward', 'cut through' and 'fragment free'. I'm implying that 'cut through' operation potentially puts those half-duplex ports into the same collision domain.

BernieM

I also consider a 'layer 3 switch' a router but for the sake of the argument was keeping it technical ie. it doesn't matter what you call the device seperating broadcast domains but 'technically' it's operating at layer 3.

This link helps explain 'subnet 0' and 'all 1's addressing':.

BernieM

Just adding to my previous post ...

If you want to take it to the lowest level .. level 1, and look at what's generating the electrical signals that 'collide' then there is no difference between a switch and (repeater) hub in regard to the collision domain. In that case each port on the hub and each 'half-duplex' port on the switch is a unique collision domain but if you're going to say the entire hub is one collision domain then doesn't the same logic apply to all the half-duplex ports on a switch ... whatever mode it's operating in.

I'm not trying to rewrite or dispute the textbooks, I'm just suggesting we look outside the square.

BernieM

No. A hub can only support one host talking at time while a switch can support many machines all talking at the same time. On a half-duplex port, the collision is between the host and the switch itself, not any other hosts. If a port on a switch is operating in full-duplex, there is no collision!

Doan

What are you saying 'no' to. I understand how hubs and switches operate. You're not getting my point. A hub doesn't buffer the transmission and repeats the electrical signal out all ports simultaneously ... hence only one host can transmit at once ... a 'true' single collision domain.

Ignoring the fact that running hosts at half-duplex on a switch is a waste of functionallity and lessens throughput ... my point is that on a switch that has multiple hosts connected at half-duplex and the switch is operating in cut-through mode ie. forwards after reading the first 6 bytes after the preamble ... which is the destination mac address, then technically speaking it temporarilly places the source host and destination host into the same collision domain because frames from each of those hosts 'can' collide.

BernieM

Ok, so what happenned when these hosts "collide"? Do they follow CSMA/CD and back off as the standard dictated?

Doan

Well the switch and destination host would but not the original sending host because it's unaware of the collision as the switch is buffering its transmission. So going through this process (explaining what I was thinking in great detail) I've come to the conclusion I was wrong. All hosts connected to half-duplex switchports are in fact in their own collision domain.

Thanks for taking the time.

BernieM

Sorry, but you're mistaken. Each half-duplex switch port always has its own collision domain, which extends between the port and its attached host. If a half-duplex port is connected to a hub, the collision domain includes the attached hub. The collision domain never extends into the switch, or between ports. "Cut-thru", "Fragment-free" and "Store-and-forward" represent implementation details, but do not change this basic fact. If a switch begins to forward a frame in cut-thru mode from one half-duplex port to another, and a collision is detected on the exit port, the collision is contained within the domain of the exit port. It is still the responsibility of the switch to re-transmit the frame from its buffer. The initiating host does not detect a collision, and will not retransmit the frame.

As for the usefulness of running a switchport at half duplex, I must also beg to differ with you. One can only run all ports at full duplex if all of your networked devices support it. Many older (and some not-so-old) devices only operate at 10/half, and require a compatible switchport to communicate. It is rarely practical in an enterprise setting (and almost never practical in a home or small business setting) to upgrade all networked components at the same time. Attaching 10/half devices to appropriately-configured switchports gives them the best possible performance and availability to the rest of the network. Just as an example, while my servers, PCs & newest laptop all run at 100/full, my own home network also includes a JetDirect printer card, Panasonic webcam, DEC termserver, HP X-server, DG supermini, two older laptops and a Cisco router, all of which operate at 10/half. Many of these devices would be considered "legacy", and as such, cannot be upgraded with newer interfaces. Others were simply designed with the lower speed because that is all they require. Attaching them all to a Catalyst switch gives them individual, isolated collision domains, and lets everything play together efficiently.