Comware Based
cancel
Showing results for 
Search instead for 
Did you mean: 

Jumbo Frames over IRF on 5820 switches

ChrisOD
Occasional Contributor

Jumbo Frames over IRF on 5820 switches

   I have a storage backbone set up on two 5820 switches connected through IRF.  The protocol i 10Gb iSCSI connected to ESXi servers to EMC VNX2500 storage arrays.

  When I constructed the switches, I set the ports to accept Jumbo frames, but did not activate them on the storage arrays or servers.  Recently a consultant advized us to turn on Jumbo frames in the situation, so I started looking at IMC to check out the proper settings on the switches.  IMC shows the physical ports that are used in the IRF as having an MTU of 1500.  The CLI does not give me an option to specify the MTU either on the physical or virtual IRF ports, so I want to confirm what is the actual setting.

   It could be that IMC does not see an MTU specified and so assumes it is the default is 1500.  How can I confirm this?

  How does IRF handle frames between two switches?  Does an MTU even matter on an IRF link?

6 REPLIES
Bjorn Lagace
Advisor

Re: Jumbo Frames over IRF on 5820 switches

Hi, I've opened two support cases already to get some answers as well. Not only do I want to know how things are handled over the IRF link.
Also two ports on the same IRF switch with jumbo frame enabled. When we test it with 2 portables (with jumbo frame configured) and we do a ping -l 9000 -f xxxx the packet gets fragmented. So I search a bit and for example default vlan1 , try to set the mtu value.....limited to 1500?!?
While on port level you have jumbo frames untill 12k.
In another forum a guy said he didnt knew the details, but knew that jumbo frames are passed, but cannot be tested with ping due to that mtu limit.

I want black on white that this switch passes jumbo frames by just typing 'jumbo frame enable'.

Ill update here if I get answers myself.
ChrisOD
Occasional Contributor

Re: Jumbo Frames over IRF on 5820 switches

I see the same symptom on VLANs - limited to mtu 1500. Procurve switch supports Jumbo frames on VLANs. I do not have the option to type 'jumbo frame enable' on the physical or virtual port on a configured IRF. Where are you doing that?
Peter_Debruyne
Honored Contributor

Re: Jumbo Frames over IRF on 5820 switches

It seems jumbo frames are supported for L2 traffic (this is what you can set on the interfaces), but not for L3 routed traffic (this is what you define on the L3 interface like int vlan x).

L2 jumbo frames are supported by default on comware devices (as far as I know), routed traffic will be fragmented.

 

This also means that jumbo frames should just pass over the IRF links if this is L2 switched traffic (but again, not in case it is routed traffic).

 

For the original question, if the storage connection is active/active with link aggregation, you should realize that no traffic will pass the IRF link at all, since IRF will prefer local-port switching if a local link is available. So all traffic entering unit1, will also leave unit1 to the destination (if destination interface is a link agg and there is a local link available on this unit1).

Only in case of link failures, the traffic would pass the IRF link.

MichaelM55
Trusted Contributor

Re: Jumbo Frames over IRF on 5820 switches


The maximum MTU value is 1500, let me explain to you why :
 
1.            MTU is normal referred to as IP MTU.
2.            Jumbo frame is Ethernet MAC layer concept to send the packet in a large size in once, because the Switch has no capability to do the IP fragment.
3.            1500 is the Ethernet packet size.
 
Jumbo frame means that if the switch receives an Ethernet packet with packet size larger than the jumbo frame size configured on the switch interface, then the Switch will discard it.
However the IP MTU is normally performed by the router (by powerful CPU), if the received IP packet size is larger than the MTU number configured on the interface, the Router will NOT discard it but fragment it.
IP fragmentation consumes a large amount of CPU resources because it has to be processed packet by packet, so it is normally processed by a Router and some high end Switches.
It cannot be done by the mid or low end Switches because they rely on the hardware chip.
 
So in our case, setting the jumbo frame to the size that will allow the packet to go through is enough, don't worry about the IP MTU



Apachez-
Trusted Contributor

Re: Jumbo Frames over IRF on 5820 switches

Here is an interresting piece of documentation regarding how Cisco handles this:

 

http://www.cisco.com/c/en/us/support/docs/switches/catalyst-6000-series-switches/24048-148.html

 

 

Catalyst 3750/3560 Series

Catalyst 3750/3560 Series switches support an MTU of 1998 bytes for all 10/100 interfaces. All Gigabit Ethernet interfaces support jumbo frames up to 9000 bytes. The default MTU and jumbo frame size is 1500 bytes. You cannot change the MTU on an individual interface. You must set the MTU globally. Reset the switch afterwards for the MTU change to take effect.
Configure

Use the system mtu command to change the MTU for all 10/100 interfaces. This command only effects 10/100 interfaces.

    3750(config)# system mtu 1546
    3750(config)# exit
    3750# reload

Use the system mtu jumbo command to change the MTU for all Gigabit Ethernet interfaces. This command only effects Gigabit Ethernet Interfaces.

    3750(config)# system mtu jumbo 9000
    3750(config)# exit
    3750# reload

Note: Gigabit Ethernet ports are not affected by the system mtu command; 10/100 ports are not affected by the system mtu jumbo command. If you do not configure the system mtu jumbo command, the setting of the system mtu command applies to all Gigabit Ethernet interfaces.

 

Unless I misunderstand someting a Cisco 3750 will support a mtu of 9000 bytes while no Comware device would support a MTU larger than 1500bytes!?

 

That sounds odd to me...

Zuntini
Occasional Visitor

Re: Jumbo Frames over IRF on 5820 switches

Same kind of question, using 6125XLG. I've set up an IRF on the four 10GbE internal crosslink, interface 17-20. The traffic is going through this IRF link as it's a data copy between two blade in the same chassis. I have set jumboframe enable on each server downlink port, on the NIC.

When I want to enable jumboframe on each IRF interface, the command option is not available. Should i break the IRF, set the jumboframe on each four interface, and then re - make the IRF ?

Or it's not necessary to do it, jumbo frame are forwarded by default through the IRF links ?

Thanks