IBM Hub/Switch Switch User Manual


 
Chapter 2 HPSS Planning
76 September 2002 HPSS Installation Guide
Release 4.5, Revision 2
The Linux raw device driver is used to bind a Linux raw character device to a block device. Any
block device may be used.
See the Linux manual page for more information on the SCSI Disk Driver, the Raw Device Driver
and the fdisk utility.
To enable the loading of the Linux native SCSI disk device, uncomment the following lines in the
.config file and follow the procedure for rebuilding your Linux kernel.
CONFIG_SCSI=y
CONFIG_BLK_DEV_SD=y
Also, depending on the type of SCSI host bus adapter (HBA) that will be used, you will need to
enable one or more of the lower level SCSI drivers. For example, if you are using one of the Adaptec
HBAs with a 7000 series chip set, uncomment the following lines in the ".config" file and follow the
procedure for rebuilding your Linux kernel.
CONFIG_SCSI_AIC7XXX=y
CONFIG_AIC7XXX_CMDS_PER_DEVICE=253
CONFIG_AIC7XXX_RESET_DELAY_MS=15000
2.6.10.4 Performance
The configuration of the storage devices (and subsequently the Movers that control them) can have
a large impact on the performance of the system because of constraints imposed by a number of
factors (e.g., device channel bandwidth, network bandwidth, processor power).
A number of conditions can influence the number of Movers configured and the specific
configuration of those Movers:
Each Mover executable is built to handle a single particular device interface (e.g., IBM
SCSI-attached 3490E/3590 drives, IBM BMUX-attached 3480/3490/3490E drives). If
multiple types of device specific interfaces are to be supported, multiple Movers must be
configured.
Each Movercurrently limitsthe number of concurrently outstanding connections. If a large
number of concurrent requests are anticipated on the drives planned for a single Mover,
the device work load should be split across multiple Movers (this is primarily an issue for
Movers that will support disk devices).
The planned device allocation should be examined to verify that the device allocated to a
single node will not overload that node's resource to the point that the full transfer rates of
the device cannot be achieved (based on the anticipated storage system usage). To off-load
a single node, some number of the devices can be allocated to other nodes, and
corresponding Movers defined on those same nodes.
In general, the connectivity between the nodes on which the Movers will run and the nodes
on which the clients will run should have an impact on the planned Mover configuration.
For TCP/IP data transfers, the only functional requirement is that routes exist between the
clients and Movers; however, the existing routes and network types will be important to
the performance of client I/O operations.