Ethernet - hubs vs. switches

G

Thread Starter

Geoff Taylor

I'm installing a network with about 12 terminals, 8 5/05's and 6 ind. pc's. Does it matter whether I go for a hub or to daisy chain and go with individual switches to avoid data collision? Is it just down to ergonomics?
 
Use switches for the number of stations you have. Hubs re-broadcast the datastream out all of the ports. Switches learn where the data is supposed to go and only broadcast out that port. With hubs, you have one large collision domain. With switches, each port is a new collision domain.

Why does this matter? Depends on the amount of traffic in your system. However, if one station goes berserk and floods its connection with continuous data or requests, it can bring the whole network to a halt if using hubs. With switches, only on or at most two connection segments will be affected.
 
It's really never an issue of ergonomics. The benefits of a switch are easily apparent. Reduced network broadcasting = faster communications. Switches only forward packets to the port the destination device is attached
to. So instead of walking into a room and everyone yelling and you try to hear a conversation, the person you're talking to walks up to you and you don't hear the other conversations. It's exponentially better for network traffic.

CK
 
K

Kirk S. Hegwood

Geoff,
Use a switched hub. It makes a world of difference in the throughput of the data. They also don't cost that much. I've recently retrofitted a PC controlled Ethernet system and the performance change was dramatic.

Good luck,

Kirk S. Hegwood
President
Signing for Hegwood Electric Service, Inc.
[email protected]

 
S

Steve St.Hilaire

All the technical mumbo-jumbo says switches are better, but my first hand experience says otherwise. We upgraded our office from 10 MBit hubs to 100 Mbit switches, and took a huge performance hit. Everything is really slow now. We used good quality 3Com 3300s. The network had 48 computers on it.

I have read info about switches having more overhead than hubs to make a connection, which may make some small file transfers slower. But there I go again with the mumbo-jumbo. I have become extremely cautious when taking others advices. You know what they say about opinions....

All I know is that I was told we would have a huge performance increase, and experienced the complete opposite. Its entirely possible that they weren't set up correctly...

I would be interested in hearing what other people think. What could I have done wrong? Does this "huge" performance increase come only if they are set up correctly? What are the common pitfalls? What is the correct procedure to get optimal performance?
 
M
Depends on size of network and availability of data requirements. Hubs will get the job done, but you do run the risks associated with creating one big collision domain. Switches are the product of choice for industrial applications since they can help make ethernet more "deterministic" and deliver critical data with greater predictability. You should also look at Hirschmann, they make a redundant self healing fiber "ring" for industrial applications. I can provide white papers aimed specifically at industrial control if interested.

Mike Finley
Industrial Networking Solutions
www.industrialnetworking.com
 
C

Curt Wuollet

The tools to find out what's going on are readily available. Even simply pinging between machines can provide a wealth of information. Chances are that switching overhead won't make a 100mbit network slower than a 10 mbit hubbed network or there would be little point in selling 100mbit switches. Something is very wrong. I suggest a
session with a packet sniffer and someone who knows TCP/IP. Just as a first guess, look to name resolution and other many to one entities. Just watching the lights will tell you if you've got a jabbering node or the like.

Regards

cww


--
Free Tools!
Machine Automation Tools (LinuxPLC) Free, Truly Open & Publicly Owned
Industrial Automation Software For Linux. mat.sourceforge.net.
Day Job: Heartland Engineering, Automation & ATE for Automotive
Rebuilders.
Consultancy: Wide Open Technologies: Moving Business & Automation to
Linux.

 
V

Vedran Kosta

Hi,
one idea cross my mind. Witch type of cable you use?
If you upgrade to 100Mb/s network, you MUST change cabling to category 5, or otherwise you will have many errors in network transmissions (looks like your case). Here is cable category specification:

EIA/TIA Category Specification provide for the following cable
transmission speeds with specifications (Note prior to Jan94
UL and Anixter developed a LEVEL system which has been dropped
or harmonized with the CATEGORY system);
Category 1 = No performance criteria
Category 2 = Rated to 1 MHz (used for telephone wiring)
Category 3 = Rated to 16 MHz (used for Ethernet 10Base-T)
Category 4 = Rated to 20 MHz (used for Token-Ring, 10Base-T)
Category 5 = Rated to 100 MHz (used for 100Base-T, 10Base-T)
UL LAN Cable Certification Program - Underwriters Laboratories
publication 200-120 30M/3/92, 1992 [characteristics of Cat 3-5 UTP]

You could simply test packet loss with PING command. I hope this information will help you,
Vedran Kosta,
"Vodovod Dubrovnik", Croatia

 
D

DAVCO Automation

This is not totally true. Each switch segment is still a collision domain (broadcast domain). If a broadcast occurs to which no specific address is
present (broadcast), the switches "address/port" table will not know where to send it, therefor all ports get it.

The way you segregate "collision domains" on a switched net is via VLANS or routers. It is true that the "addressed"messages are passed only between the communicating ports........this is the functions of a level 2/level three device.

Dave
 
B
Does the exisiting ethernet cabling have both pairs in it? 2 pairs are required for 100BaseT while only one pair is used for 10BaseT. maybe the switches sensed only one pair and defaulted to 10baset or something.

Most switches these days will shut off a port that has some kind of problem on it so its unlikely that a single problem port could create such an issue. Check the lights on your switches for error indications. You can probably
connect to the switch via IE or a terminal and read what errors there might be.
 
As I stated in another reply on this topic, the tools you can use are in the IOS software in the CISCO switch. In case you didn't all know, this is
probably the 2nd leading OS in the world if not the 1st.

Find someone who knows CISCO switches and thier OP systems (IOS), as stated earlier....a 1 week class just on switches and only scratches the
surface.........a whole lot to know about MAC tables and running-congig commands etc......

So now if you would like you can start a whole new thread on the evils of CISCO's IOS op system and its proprietary nature............

No pun intended........and no evil either :) :| :(

Dave

DAVCO Automation
"The Developing Automation Value Company"
 
I am not a CAT 5 10Base T guru but I believe that 1/2 duples e-nat uses only 2 wires and full duplex uses 4.....but correct me I am wrong :) :| :(

Dave

DAVCO Automation
"The Developing Automation Value Company"


 
Assume they are set up wrong, there is no way you should experience a decrease in performance. All the reasons and setup of Cisco switches are
hard to describe here without going into technical "mumbo jumbo", but this is a perfect example of why a Cisco CCNA costs thousands of dollars and time to attain.

As example, I just took a Cisco course (don't have the number in front of me) which just covers switches, that was 1 week long and just really
scratched the surface of the deptghs that this topic covers.

Without getting some running-config files from you and looking at the configuration cannot say.

One caviet, I am using a different model so your mileage may vary.

Dave
 
B

Bruce Jorgensen

We upgraded to 100Mbps using an existing Cat 5 cabling installation. We had lots of problems. We then rewired with Cat 6 and all problems were resolved.

Cat 5 is dead since 1998 per Lucent and 5e since 2000.

Other point. Note that the cable pairing is different for 100Mbps compared to 10Mbps. Are you still using a lot of 10Mbps cables. This was also found to be a problem. 100Mbps cables work fine for 10Mbps. The inverse is not true.

Bruce Jorgensen
www.interlog.cl
 
>...snip
>Other point. Note that the cable pairing is different for 100Mbps
>compared to 10Mbps. Are you still using a lot of 10Mbps cables. This >was
>also found to be a problem.
>...snip

NO it is not. The cable pairings are the same for 10Mbps and 100Mbps. The pinouts are pins 1,2,3 and 6 on the RJ45 connector. This is true for 10Mbps and 100Mbps.

Pin 1: TX+ |one cable
Pin 2: TX- |pair

Pin 3: RX+ |second cable
Pin 6: RX- |pair

Make sure the cable pairs for TX and RX stay together. If the TX+ and TX- or RX+ and RX- are not in the same cable pair poor perfermance at 100Mbps speeds will be the result.

Bradley G. Hite
Intertech Incorporated
mailto:[email protected]
http://www.myplc.comTeaching Practical Skills for a Technological World

 
D

Daniel M. Kamers

Standard Cat5 is OK for 100Mbps.
Cat5e its just horizon (Gigabit Ethernet and full-duplex). Cat6 is not yet an aproved standard (I think so).

There are at least two common types of wiring for Ethernet. They are not compatible.

After all, 10 mbps uses 2 pairs, 100mpbs uses 4 pairs, but they are compatible if the same type of wiring is used.

Never let anyone do an engineer's job...or do it yourself.
Best regards.
----
Daniel M. Kamers
 
> There are at least two common types of wiring for Ethernet.
> They are not compatible.
>
> After all, 10 mbps uses 2 pairs, 100mpbs uses 4 pairs, but they are compatible if
> the same type of wiring is used.

NOOOOO. Again 10Mbps and 100Mbps only use two (2) pairs! There was an older pre-standard version of 100Mbps Ethernet that did use four pairs but that
pre-standard version (was it 100Vfast?) is as far as I know a dead product.

You can prove this to yourself if you want by splitting out one (1) cat5 4 pair cable into two (2) 10Mbps or two (2) 100Mbps connections.

See my previous post for pair utilization and pinouts of RJ45.

Bradley G. Hite
Intertech Inc.
mailto:[email protected]
http://www.myplc.comTeaching Practical Skills for a Technological World
 
J

Johan Bengtsson

Actually the packets will get out to all ports - that is true but there will not be collisions from it, they will simply be sent after each other. When you have collisions in a non switched network is when two devices try to send at the same time.

I don't know if this might be the difference betwen a switch and a switching hub however (could someone enlighten me here because I have wondered about that for some time now).

For a real switch a message will be sent to the port where the intended reciever is, if it is a broadcast message it will be sent to all ports except the one it came from. If two messages want to be sent out thru the same port at the
same time one will be delayed until the other one is sent.

My guess (I do *not* know this) is that a switching hub will do the same but without the possibility to delay outbound messages. Can someone deny or confirm this?

/Johan Bengtsson

----------------------------------------
P&L, Innovation in training
Box 252, S-281 23 H{ssleholm SWEDEN
Tel: +46 451 49 460, Fax: +46 451 89 833
E-mail: [email protected]
Internet: http://www.pol.se/
----------------------------------------

 
Top