HP 3PAR Ubuntu Operating System Implementation Guide

/etc/rc1.d/S41open-iscsi
/etc/rc6.d/S41open-iscsi
/etc/rcS.d/S25open-iscsi
6. To prevent one host from using up all the I/O buffers, you will need to throttle the host I/O
by reducing the node.session.cmds_max and node.session.queue_depth values.
The following values were determined based on the multihost-to-single HP 3PAR StoreServ
Storage port testing completed at the HP test labs and may require further tuning for your
particular configuration.
node.session.cmds_max = 32
node.session.queue_depth = 8
CAUTION: The HP 3PAR StoreServ Storage target iSCSI port has a limited number of I/O
buffers and, with many iSCSI initiators connected to a given port, will result in queuing that
exceeds the I/O buffer capacity and could lead to network and I/O timeout issues, as well
as application failures.
7. Session and device queue depth in /etc/iscsi/iscsid.conf may require tuning
depending on your particular configuration. node.session.cmds_max controls how many
commands the session will queue. node.session.queue_depth controls the device's
queue depth.
If you are deploying the HP 3PAR Priority Optimization software, you may need to increase
or max out the node.session.cmds_max and node.session.queue_depth values to
ensure the host has sufficient I/O throughput to support this feature. For complete details of
how to use Priority Optimization (Quality of Service) on HP 3PAR StoreServ Storage arrays,
please read the HP 3PAR Priority Optimization technical white paper available on the following
website:
HP 3PAR Priority Optimization
In a multihost-to-single HP 3PAR StoreServ Storage configuration when HP 3PAR Priority
Optimization is not in use it is possible to overrun the array target port I/O queues or
experience queue starvation for some hosts due to excessive usage by other hosts. These
situations can be mitigated by reducing the values for parameters node.session.cmds_max
and node.session.queue_depth on each host that shares the array target port.
8. As an option, you can also enable the Header and Data Digest for error handling and recovery
within the connection.
Typically, whenever a CRC error occurs, the SCSI layer tries to recover by disabling the
connection and recovering. However, by enabling the header and data digest, individual
iSCSI PDUs will be retried for recovery for those connections missing the data (CRC error) or
missing a PDU or sequence number (header digest). If the recovery does not occur, then the
low level SCSI recovery will be initiated. The header and data digest is optional, because the
SCSI layer will still perform CRC error recovery at the session level rather than at the PDU
level.
CAUTION: Enabling Header and Data Digest will cause some I/O performance degradation
due to data checking.
You can enable the Header and Data Digest by adding the following lines in iSCSI
configuration file /etc/iscsi/iscsid.conf:
node.conn[0].iscsi.HeaderDigest = CRC32C
node.conn[0].iscsi.DataDigest = CRC32C
12 Configuring the HP 3PAR StoreServ Storage for iSCSI