HP StorageWorks Scalable File Share Release Notes - Version 2.2-1

Known issues and workarounds in HP SFS Version 2.2-1 1–9
1.3.3 Configuring the subnet mask on a HP SFS system with an InfiniBand
Interconnect
If you intend to attach your HP SFS system to a HP XC system using the InfiniBand Interconnect you must
ensure that the subnet mask associated with this network is identical on both systems. For example, if the
subnet mask of the InfiniBand interface on your HP SFS system is 255.255.0.0, then you must also configure
your HP XC system to use 255.255.0.0 as the subnet mask of the InfiniBand interface.
1.3.4 Voltaire InfiniBand stack fails to load if the node chipset does not support
MSIX
If the Voltaire InfiniBand stack fails to load with error messages such as the following, then it is likely that
the node chipset does not support MSIX:
THH(1): /mlxhh/thh/cmdif_comm.c[864]: Command not completed after timeout:
cmd=TAVOR_IF_CMD_MAD_IFC (0x24), token=0x1300, pid=0xBE0, go=0
You can check if the node chipset supports MSIX as follows:
# grep -i msi /proc/iomem
If the node chipset supports MSIX, the output should be similar to:
fdf82000-fdf821ff : MSI-X vector table
If the node chipset does not support MSIX, then no output is returned. You should add the following to the
/etc/modprobe.conf text file (or any more relevant modprobe.conf.* text file):
options mod_thh disable_msix=1
options mod_rhh disable_msix=1
1.3.5 Tuning OFED o2ib LND performance for MT23108 HCAs
The default OFED 2048 byte MTU setting for MT23108 TAVOR HCA adaptors is not optimum for Lustre
performance. Performance may be degraded as much as 50% if any of the two endpoint HCAs are
MT23108 HCAs. To overcome this defect, the MTU setting must be overwritten by adding a line such as the
following in the /etc/modprobe.conf.lustre text file on the node that establishes the Lustre
connection:
options ko2iblnd ib_mtu=1024
This is the case for HP SFS client nodes and HP SFS MDS nodes and must be configured if the local or
remote, or both node HCAs are MT23108 HCAs.
You do not need to configure this setting on the HP SFS server side, it is performed automatically. If you use
the sfsconfig utility, you do not need to configure this setting on the client nodes, the sfsconfig utility
sets it up automatically by probing both HCAs.
At Lustre start time, kernel messages such as the following indicate that the default MTU was overwritten:
kernel: Lustre: 2912:0:(o2iblnd.c:737:kiblnd_create_conn()) current IB mtu 2048,
requested 1024
1.3.6 OSS on HP ProLiant DL360 G4 servers should be rebooted after upgrade
After upgrade from HP SFS Version 2.1 to 2.2, the status of arrays connected to Object Storage Servers may
be replaced (instead of online). This happens on HP ProLiant DL360 G4 servers when they are first
booted.
The solution is to shut down and then reboot all Object Storage Servers. The status of the arrays changes to
online when the servers reboot.