Dell Emulex Family of Adapters - Manual (Page 667)

Emulex Drivers for Windows User Manual P010077-01A Rev. A
3. Configuration
iSCSI Driver Configuration
667
Maximum Transmission Unit (MTU) for iSCSI Connections
Because the Emulex OneConnect UCNA is a multi-function adapter, the MTU settings
for iSCSI functions are different than the ones for NIC functions.
For iSCSI, there is no explicit way to configure MTU from the OneCommand Manager
application. Instead, this value is auto-negotiated by the firmware. Before establishing a
TCP connection for an iSCSI Login, the iSCSI firmware issues an ICMP Echo with a
large payload to the iSCSI target. If Jumbo Frames has been enabled on all the switches
leading to the target, as well as, on the target interface and if there is a successful ICMP
Echo reply, the iSCSI firmware uses Jumbo Frames for that connection. The MTU used
in this case is 8342 bytes.
If the large ping request is unsuccessful, the firmware defaults to non-jumbo mode and
uses an MTU size of 1514 bytes.
The Max MTU value is displayed in the OneCommand Manager application for the
iSCSI controller under the Port Information Tab on the Max MTU field. The TCP MSS
used for an active iSCSI connection is displayed in the OneCommand Manager
application in the ‘TargetSessions’ screen on the TCPMSS field.
iSCSI Error Handling
The goal of iSCSI error handling is to be tolerant of link level and target level failures
up to configured timeout values, so that I/O errors are not seen by the application or
operating system. The error handling is triggered under the following conditions:

Loss of immediate link to the UCNA (such as a cable disconnect or port failure).
The UCNA firmware detects the loss of link and notifies the driver. When this
happens, the driver queues the I/O requests internally, up to a configured
timeout period, so that the operating system does not see I/O errors. This
timeout period is known as LDTO.

Loss of connection to the target because of target or network disconnection at
the target. If the driver has I/O requests pending with the target and the target
becomes unavailable (because the target is down, has failed over, or network
issues are detected at the target), the driver queues the I/O request internally
up to a configured timeout period. This timeout period is known as ETO.
If the configured threshold for LDTO and ETO is reached and the UCNA is still unable
to connect to the target, the driver fails all I/O requests. I/O errors are seen by the
application and operating system.
Note: Following a link up, switch ports can take a long time to initialize and go to a
forwarding state. Because of this, add additional time to the ETO and LDTO
settings to eliminate I/O disruption or target unavailability. If the switch port is
connected to a single host, then PortFast mode can be enabled on the switch
port to eliminate delays in transitioning to a forwarding state.