MTU

[singh.yashpal]

The default Maximum Transmission Unit (MTU) size for frames received and transmitted on all interfaces on the switch stack is 1500 bytes , on a gigabit interface v can increase the MTU from 1500 bytes

upto 9000 bytes by using System jumbo mtu command .

My question : Is the increase in MTU size( 06 times) can have a significant change in the performance

I have got one more query to add in it.

Why is the minimum frame length 64 bytes in case of 10/100 Mbps Ethernet while it is 416 & 520 bytes for 1000 base X and 1000 base T respectively?

[lgijssel]

Increasing the MTU reduces the percentage of overhead that a datagram contains. From that prospect of view: the larger, the better.

The max mtu-size merely serves to keep the bus accessible for all nodes. One node cannot claim the network for an unlimited amount of time.

The minimum sizes are related to timing issues on the physical layer. If the frame is smaller than this, it will not be possible to detect collisions in a reliable way.

Regards,

Leo

[bhedlund]

You can increase the MTU setting on your router/switch interfaces. However it wont do you much good unless your servers are using jumbo frames as well, and even then it would only help server-to-server within the LAN. To see the real benefit you would need the workstation/desktops configured to use jumbo frames with the same LAN environment as your jumbo frame servers.

[rajeshiyer]

Hi Techies,

I am new to Networks. Just learning to grab some T/S info from this forum.

Can you pls explain me the term "Overhead". I hear this very often. couldn't manage to get details.

Thanks in advance.

Rajesh Iyer

[lgijssel]

Hello Rajesh,

Isn't there anybody at IBM that is able to explain this to you? ;-) I suppose they have more important things to do.

Overhead is all extra data that is added while the data traverses down the OSI layer. TCP adds a 20 byte tcp-header and IP also adds 20 bytes at least.

In an extreme case, although it happens very often, a one-byte keystroke requires 64 bytes when it traverses an old-fashioned ethernet network.

So this "overhead" is not useful to the enduser but is still needed to transport the data in a standard way. I hope that this answers your question.

Regards,

Leo

[rajeshiyer]

Hi Leo,

Your explanation is very much appreciated !

Thanks,

Rajesh Iyer.

[lgijssel]

Thank you for rating all helpfull posts. ;-)

Leo

[gnijs]

Why is the minimum frame length 64 bytes in case of 10/100 Mbps Ethernet while it is 416 & 520 bytes for 1000 base X and 1000 base T respective ??

-----------------

Now this is a good question. Why is their something like a "minimum frame length".

Actually this is quiet logical as long as you remember that transmitting a bit over a wire from one end to the other takes some time, not very much, but some time anyway.

Classical Ethernet is based on detecting collisions. So suppose two workstations are located on each end of a long Ethernet cable (100m). IF station A transmits something on the wire, it must transmit LONG enough so that it is sure that his signal has reached the complete length of the cable. IF it was only transmitting one bit for a very short time, the two PC stations COULD transmit something VERY short at the same time without being able to detect that they are transmitting because the signal has not yet reached or travelled to the other PC.

That is why the maximum length of a cable and the minimum frame size are related.

For FastEthernet is seems that 64 bytes is long enough to travel the whole 100 meters.

Also when using gig ethernet, the NIC is putting the bits +/- 10x times faster on the cable, so you have two choices: either the max cable length shortens by +/- 10 times to 10 meter (unacceptable) or you increase the minimum time a NIC has to send by +/- 10 times. This is what the IEEE standards people had to do.

Geert

[dnewell24]

Geert,

I like your explanation but I’m not sure I agree with everything.

True with gigabit ethernet, the NIC is putting the bits 10x times faster on the cable. As result the bit-time, or time to recognize a collision in half duplex operation, is 1/10th the size of Fast Ethernet; resulting in a max network diameter of 20 meters. To achieve the 200 meter network diameter goal IEEE engineers “extended” the frame size. IEEE added a carrier extension (non-data) at the end of the frame extending to 512 bytes. Any frame less than 4,096 bits or 512 bytes are appended with a carrier extension.

This does add overhead so Frame Bursting was adding to the standard to increase efficiency.

Ryan