Ethernet 40% bandwdith Rule

[RebelFox]

Hi,

I am studying for a Network + qualification and one of the topics covered is the Ethernet standard. I have been told that a fact of Ethernet is that it uses 40% of the available bandwidth. Can anybody tell me why it is so inefficient and what the other 60% is used for? My lecturer belives it is used in some way to handle collisions but he is not sure how.

If anybody could explain to me what the missing 60% is used for and how it is used I'd be grateful.

 

[franklin97355]

Read the second box from the bottom of this page, it applies to you. But I'm sure someone here might have an answer. I don't.

 

[edfair]

Good question. If the instructor can't answer it, it sounds like it really isn't a rule.
At 40% utilization I would suspect it would go to 100% pretty fast with the retries being forced by collisions.
You have 2 machines. Dump 2 good size files each way singly and time them, Then dump them across the network at the same time and see if they both finish in the combined time.
Have never tried it, so you can help me out by proving or disproving my theory that both being used at full transfer speed will slow the network down. And 2 accessing may not be enough to bring the network to saturation.

Ed Fair
Give the wrong symptoms, get the wrong solutions.

 

[jimbopalmer]

When ethernet was all co-ax cable, only one device could send at a time and every packet went to every device. Hubs are essentially the same, only one device can send at a time and every packet went to every device. This means after each packet every device that wants to send a packet tries to send, until one does not collide. this can use up a lot of bandwidth.

With switches, devices can send and receive at once and packets only go to devices that need to see them, as a result, very few collisions occur. As well, switches can have buffers to reduce collisions further, deferring packets rather than colliding.

Full duplex switched ethernet can have almost 100% activity, and is cheap enough you almosst never have a reason not to use it.

I tried to remain child-like, all I acheived was childish.

 

[RebelFox]

franklin-

The student rule refers to assistance with homework.
That is not what my question is about. Its extra understanding. I'm also an Analyst/Programmer so 95% of my postings relate to that and not any extra qualifications i am chasing.

Clarifaction anybody if I am breaking any rules?


Thanks for your help though.

 

[edfair]

No, you are not breaking any rules.

Ed Fair
Give the wrong symptoms, get the wrong solutions.

 

[franklin97355]

Sorry, It just sounded like homework from the way it was worded. Like Ed says , it's a good question and has started several people thinking.

 

[1Drisnil]

I think Jimbopalmer has something here. You need to recieve additional parameters from your instructor. Is the ethernet in a pure non-switched environment? What cabling is used?

At work we have a full duplex Fast Ethernet Switched network...never seem to have issues with collisions.

Hope this helps you understand the question a bit better.



"...understanding "The Why" is half the battle!" - Anonymous.

 

[rwullems]

Hi,

This rule is getting very outdated. It was (or still is) used for a shared enviroment (like a hub or coax segment) and is ONLY appicable for Half Duplex.

The base of the rules is that is you are increasing the utilization, this will increase the amount of collissions, wich will increase the amount of retransmissions, wicht on his turn wil increase the utilization.

This is a cricle where you get in, and your utilization will keep increasing while your effective troughput will go down caused by all the collisions.

As Jimbopalmer and some others already mentioned, this is not applicalbe to an Full duplex enviroment, where collisions DO NOT exist.
regards,
Robert

Robert A.H. Wullems
Sniffer University Instructor
SCM/CNX/MSCA

 

[Grenage]

If it helps at all:

On our 100mb switched network the average throughput with "real" data is around 7MB a second (70% utilization).

It would most likely be slightly higher but many of the existing cables are of poor quality.

 

[GTomte]

In a lab, with only two devices, you can push data from one client to a receiver with 100% utilization. But then the receiver can't send out any packets.

But as the people here already has said:
- The 40% rule only happens with half duplex
- Hub's are always half duplex
- With full duplex, there will not occur any collisions

I have worked with a network that had 40-45% load, and it was extremly slow.

Another thing in networking, is this:
- Mismatch on duplex!!

The server is configured with 100 Mbit full duplex, the switch port is auto, and senses 100 Mbit half duplex. This is a correct behavior. What do you think the utilization will be???

- 1-1,5 Mbit/s. Slower then a 10 Mbit/s hub.

Configure your network correct.

Regards
G. Tomte

- Nortel Networks, Routers, switches
- Alteon
- CacheFlow
- Blue Coat

 

[bcastner]

What bothers me is that rather than the "truth" for the exam you better memorize the answer expected.

My problem is the notion of bandwidth utilitzation. Others have pointed out the relevance of this in half-duplex settings, but the general question seems flawed to me.

There are several vendors offering 1 gig ethernet over copper. The bandwidth should refer to the ability of the copper medium to carry traffic. The "standard" ethernet specs for 10 and 100 traffic likely use a small portion of the possible bandwidth of their copper connections.

 

[rwullems]

Yes,
You have a point there, but if you are gonna look into that, that has more to do with the way the signal (the bit stream) is coded. There are different coding schemes like 4B5B, 8B10B and so on. Gigabit for example is using 8B10B (over copper, 802.3ab). So if you need to sent a byte (8 bits), you need 10 bits on the cable. So this is generating not 1Gbps on the cable, but actually 1.25Gbps!

Even so, the difference between 10Mb and 100Mb is very small. The bits are sent out ten times faster wich will make the bit 10 times smaller in length (for example, a bit on a 100Mb UTP network has a length of 1.77 meters)

Probably it is possible the use a higher bandwith on copper, but the biggest limitation there will be the size of the network. A bit at gigabit is .177 meters, If it is possible for example to create a 10gb link over copper, the bit will be .0177 meters, wich will decrease the size of your network with a factor 10. The limitation is generated by the smallest frame size on Ethernet, that is 64 bytes = 512 bits. So calculating the round trip time on a single cable = 512/2 = 256 bits * 0.0177 meters, you network cable should theoritical not be longer then 4.5 meters!!

Hope this helps you out a little.
regards,
Robert

p.s. Happy New Year

 

[bcastner]

Robert,

Thank you, and Happy New Year to you as well.

You have explained well why 1 gig. over copper is expensive to do. 10/100 does not require such an expensive investment in actively repeating (and timing) the traffic over any distance.

It just struck me as stated by the original poster as a remarkably bald statement about the inefficiency of Ethernet. Having gone through thick ethernet, thin-net ethernet, ARCNET, and IBM Token Ring, I just have a knee-jerk reaction to defend ethernet over copper twisted pairs.

The bandwidth possibilites under Ethernet with copper are much larger than suggested by the original poster. But, as you noted, it will not be cheap to utilize this over any distance.