MTU between switches

[yyagolyyagol]

Hi All ,

Its my first post and im no CISCO expert , need some understanding of MTU .

I have this cenario :

2X 2960 switches connected with 1000FD trunk interfaces .

the MTU is set with default 1500 .

if i set the connection MTU to 4500 between this switches , and have other interfaces set with

default 1500 , what would happen ? should i expect for packet losts or would it be OK .

the reason i want to have MTU 4500 is that MTU of 1500 maximum speed cannot get over 32Mb/s

and there for cannot use the full speen of the 1Gb ethernet/interface .

[gatlin007]

The path MTU is the smallest MTU on any interface in the path between two hosts.  So if any interface MTU is 1500 bytes regardless of how large the other interfaces MTU is; the path MTU will be 1500 bytes.

If a frame is sent that exceeds the MTU of an interface it is discarded.

The MTU for a switch is usually set globally for the entire switch.  That said; this is code and hardware dependant.  Please get the configuration documentation for your exact hardware/software versions.


Chris

[Mahesh Gohil]

Hi yan,

Nothing will change..you will face no packet loss. But yes it is always good practice to have consistent MTU between

all networking devices. reason is you face problem if

1- Your device is sending packets with df bit set and there is device inbetween whose mtu is less than the packet size.

2- If you have l2 mpls link somewhere and it is passing through this low mtu node.

anyway choice is yours it is better to have consistent mtu .

By the way i didn't understand below statement

i want to have MTU 4500 is that MTU of 1500 maximum speed cannot get over 32Mb/s
and there for cannot use the full speen of the 1Gb ethernet/interface 

Regards

Mahesh

[yyagolyyagol]

Thanks for the reply ,

I understand now that according to the chema i drow

if WEB1 sends packets to WEB2 i should not expect for any packets drop

I do need to read if all devices (firewalls,load-balances ..etc. ) can be set with the 4500 MTU

so i can set all LAN with same MTU for best performance .

the documnet about MTU i read was this one

http://www.apparentnetworks.com/pdf/white_papers/whitepaper_jumbomtu_sep2003.pdf

[antasson]

Hi Yan,

the best practice is to configure on the edge ports (toward the end users) 1500 and in the core-facing interfaces a bigger MTU to accomodate for any possible encapsulation header you are or will use in your network (QinQ for example).

Bottom line, you shouldn't hit any problem with it.

Regards,

Antonio

P.S. I also didn't understand the statement regarding the 32Mbps maximum with 1500 bytes between the switches

[yyagolyyagol]

Thanks

I read the user documents for Catlyst 2960 and it does support jumbo MTU up to 9000
also the servers interfaces NetXtreme II BCM5708 Gigabit Ethernet suppport up to 9000.
the " 32Mbps maximum with 1500 bytes between the switches" i regard to MTU 1500

that the maximum performance can get up to 32Mb/s max , i tested that whith moving large files

between servers , and never got more then that . thats why i start investigating the reason

and got to this documents explaining that jubo MTU may help here

http://www.apparentnetworks.com/pdf/white_papers/whitepaper_jumbomtu_sep2003.pdf

[antasson]

Hi Yan,

did you observe any packet drop when doing the tests with 1500 bytes in the core?

Seems very strange to me...

Antonio

[yyagolyyagol]

Well ,

I maybe wrong here , the tests i did was over scp . i guess

scp got speed limitation . white testing over rsync i got a much fater results

[root@XXX tmp]# rsync -av rsync://10.11.12.100/pub/1G.file .

receiving file list ... done

1G.file

sent 97 bytes  received 1048704116 bytes  83896337.04 bytes/sec

total size is 1048576000  speedup is 1.00

over FTP i got up to 58Mbyte/s

but rsync is 80Mbyte/s .

i guess we can close this post , i will sty with MTU of 1500 for now

since this is a production area that i dont want to mess around with .

thanks for everyone who helped this is a great froms and pleople i can only learn from .