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

New and changed features in HP SFS Version 2.2-11–6
1.2.15 Interconnect is determined by LNET options only
In releases prior to HP SFS Version 2.2-0, the interconnect used for Lustre traffic was specified as an option
to the sfsmount or lconf commands (and also in /etc/sfstab or /etc/ststab.proto). Starting
with HP SFS Version 2.2-0, this is no longer the case. The selection of the interconnect is dictated by the
LNET module network option as described in Appendix A of the HP StorageWorks Scalable File Share Client
Installation and User Guide and Chapter 7 of the HP StorageWorks Scalable File Share System User Guide.
When several possible interconnects are available, the system selects an interconnect as follows:
• Lustre selects the first LNET specified in the network option that is shared by both the client host and
the HP SFS server.
• Once a client has mounted a file system, all subsequent file system mounts on that client from the same
HP SFS server use the same interconnect. This feature is inherent in the design of LNET and differs
from the behaviour of all previous versions of HP SFS.
When you specify a network while mounting a file system, you are not specifying the network to be used
for Lustre traffic. Instead, you are specifying the network used by the mount command to contact and retrieve
configuration data from the MDS service.
For example, the following command specifies that the tcp0 network should be used to contact the mds3
service:
# sfsmount lnet://16.123.123.1@tcp0:/data-mds3......
This causes Lustre to use tcp0 to retrieve file system configuration information from 16.123.123.1 on the tcp0
network. However, once the configuration data is retrieved, the mount command uses the LNET network
options to determine which interconnect to use. If both the client and HP SFS server lists o2ib0 before tcp0
in their LNET network options, then the o2ib0 network is used for Lustre traffic—not tcp0.
In HP SFS Version 2.2-1, the sfsmount command checks if the interconnect actually used by Lustre is the
interconnect requested on the sfsmount command line. If not, the sfsmount command displays and logs
the following warning:
<filesystem> may actually have been mounted over <lnet> instead of <lnet>
Seeing this warning means that the mount succeeded, but that the performance may not be as good as
expected over the requested LNET. Because it not always possible to detect exactly which interconnect Lustre
is using, on receipt of this warning you should check which interconnects are actually in use.
The legacy sfslconf and lconf commands are also affected in the same way. The -net or -nal option
do not determine the interconnect used.
1.2.16 HP Cluster Platform InfiniBand Fabric Management and Diagnostic Guide
is now included on the HP SFS CD-ROM
Detailed instructions for upgrading firmware on HCA adapters can be found in the HP Cluster Platform
InfiniBand Fabric Management and Diagnostic Guide. In HP SFS Version 2.2-1, the HP Cluster Platform
InfiniBand Fabric Management and Diagnostic Guide is included on the HP StorageWorks Scalable File
Share System Software CD-ROM.