Cascade Hubs Switches 1

[iolair]

I was reading some other messages in this forum, and am wondering why cascading hubs or switches is considered bad design? I have a 10BaseT network that works fine, but someone donated a 100BaseT switch to us, and I am wondering if I could use that as a backbone to the other hubs/switches? We use Netware as our OS. Thanks.

Iolair

 

[daronwilson]

The 5-4-3 rule is one designed for a single collision domain, and when you are operating a 10bT network with hubs (no switching) you are dealing with one collision domain. Here is an explanation:


Collisions are an inherent part of Ethernet. Since a sending station may transmit at any time, its transmission may collide with transmissions from other stations. When a collision is detected, the sending station backs off, waits awhile and attempts to retransmit. This retransmission process is very efficient when performed by hardware. In a large network, if the data travel delay time between two stations is too long, a collision may occur without the sending station's hardware detecting it. Typically, a software function detects the loss of the sent data and initiates a retransmission process. This retransmission process is very inefficient when performed by software.

The 5-4-3 collision rule was defined by the IEEE 802.3 standard as a set of criteria that when followed, will ensure the detection of collisions by hardware. The following paragraphs summarize the main criteria points.

The main criterion dictates that two stations (DTEs) should not be separated by more than 5 segments and 4 repeaters. In the maximum configuration the following restrictions apply: (a) No more than 3 of the segments should be of a shared type (e.g. coax). (b) No fiber segments should exceed 500 m.

When two stations are separated by 4 segments and 3 repeaters the following restrictions apply: (a) Inter-repeater fiber segments should not exceed 1000 m. (b) Station to repeater segments should not exceed 400 m. There are no restrictions on using shared segments.


As the switch functions differently, and effectively creates seperate collision domains, it would probably serve you well as a device to seperate collision domains and feed the various hubs out of ports on your switch.

Here is another simple explanation:


All computers connected by the same length of coax, or by hubs/repeaters, are said to be in the same collision domain. That is, if any two of them transmit at the same time, a collision will occur.

If the collision domain is too large, it would be possible for two machines to transmit at the same time and to finish their transmision before they heard the other transmision. A collision would have occured, but neither of them would know so they wouldn't retransmit. Obviously this is not a good situation, hence the 5-4-3 rule.


So yes, using your switch at the backbone and feeding the existing hubs would be a good improvement. Also, getting the server(s) their own ports on the switch would make things smoother as well.

Hope that helps

It is only my opinion, based on my experience and education...I am always willing to learn, educate me!
Daron J. Wilson, RCDD
daron.wilson@lhmorris.com

 

[iolair]

Mr. Wilson,
Thank you for your reply. YOur explanation was crystal clear and very helpful. I now know how I can proceed. Yes, I will have the opportunity to also get the servers on their own ports as well.

 

[Olivcrom]

I was wondering how many switches you can cascade before you encounter problems? Its said that switches are not covered by the 5-4-3 rule... So, how many can I cascade? I've tried 6 segments and 5 switches and it worked... Just wondering what my theoretical and actual limits are...

 

[iolair]

Olivcrom,
Interesting question. I'd like to know the answer as well. Not that I'd need it here, as I have only four segments, but it would be nice to know. We may expand one day...............

Iolair

 

[jimbopalmer]

Ethernet expect a Collision to occur near the front of the packet. If a collision occurs 'late' then it is a sign that the network has too much latency. As your packets go though the switches it gets later and later: in effect the switches slow the packets, they get all the bits before they start tansmitting any bits.

If you start seeing 'late collisions' your subnet is too large, physical distance and perfomance of each switch will make it hard to give a firm rule about how many is too many. I tried to remain child-like, all I acheived was childish.

 

[Olivcrom]

Dumb question... He he... How do I find out if there is a late collission occuring? Will it show up as collission in my switch? LED blinks it there is a collision, apparently...

 

[jimbopalmer]

Olivcrom on a managed switches (telnet, SMNP, Web interface, or vendor software) you can get those statistics. On unmanaged gear, the blinkey lights are all you have, unless you bring in protocol analyzers. Sorry. I tried to remain child-like, all I acheived was childish.