EqualLogic

[ssreereddy]

Hi everybody,

can u please tell me how can i change the MTU setting in my ps6500 storage box.  Currently MTU size 9000 bytes.  i want to modify.

Thanks in Advance

SSeereddy.

[DELL-Joe S]

The EqualLogic arrays support standard Ethernet frames and Jumbo Frames up to 9000 bytes (including the TCP header), and will detect path MTU, which is the ability to automatically detect the maximum frame size between TCP/IP end point.  Jumbo size is determined by the negotiation between the end points.

For replication, starting in Firmware v4.1, a new support-only command exists to persistently stop using jumbo frames for outbound replication traffic. You run this on the source group, and it applies to all outgoing replications.

GroupName> support repl-use-jumbos [yes|no]

-Joe

[Anonymous]

To add to what Joe said, this is done on a session by session basis. You can mix standard and jumbo frames together in the same environment.   If the same host is using both, then suspect a configuration problem at your switch.

Regards,

[ssreereddy]

Hi Joe & don,

Thanks for your value reply.  Is there any option to modify 9000 bytes?

Thanks

sseereddy

[Anonymous]

No, there's no option to change it, except for replication as Joe noted.  Why do you think you need to change it?

[kristofvm]

I have a customer using HP Procurve 28xx series for iSCSI which are known for issues with flow-control as well as jumbo frames enabled at the same time... http://communities.vmware.com/thread/268078 . They now bougth a PS4100E they want to use for VMFS volumes.

When using flow-contol, I prefer not to use jumbo frames. Therefore, I woul like to set at EQL level. At ESXi 5.0 I keep the MTU at 1500 bytes..

Once iSCSI configured, I also only see one path, although RR is set on the datastore. In the vmkernel port binding I can see one still as last active. the other unused... Why not both as active which I am used to seeing?

Is it possible to hardcode MTU on the EQL for normal traffic just like replication traffic?

Kind regards.

[Anonymous]

You can't disable jumbo frames on the array.  However, if the physical switch and ESX vSwitches are set for standard frames then Jumbo Frames will never be negotiated.  

Re: Round Robin.  Sounds like it's not configured correctly.  

From the ESX console please copy the output of:

#esxcli storage nmp device list

# esxcli iscsi networkportal list

#esxcfg-vswitch -l

#esxcfg-route -l

#esxcfg-vmknic -l

[joergriether]

to clarify: even if your san is set to mtu 9000 and the switchports are also set to mtu >=9000, IF THE ESXi is set to 1500, 1500 it is all along the way!

Best regards,

Joerg

[Anonymous]

that is correct, however, if you should add another array to the group (or on the same switch), the groups will communicate via Jumbo Frames.

On the HP 28xx, with the latest FW, you can set the buffers to a single Queue.  By default the buffer memory is split into for QoS Queues.   Changing to one will improve iSCSI performance significantly.

There was a flowcontrol issue resolved in HP's f/w version I.08.61, however, they have addressed a great deal of other issues since then, so you probably should ensure you are at a recent vintage, and if not at the latest, review their release notes to ensure any newer fixes aren't important to you.

Another setting we recommend if you are not using qos, is to enable qos-passthrough-mode. This will provide more buffers for iSCSI traffic, which are important to provide the best throughput.

Support HP Procurve 2848 1# config

Support HP Procurve 2848 1(config)# qos-passthrough-mode one-queue

Support HP Procurve 2848 1(config)# exit

Support HP Procurve 2848 1# write memory

Support HP Procurve 2848 1# exit

Note that this change will require a restart of the switch, which will take at least two minutes. Do it only during a maintenance window.