Tools and Utilities - Expert Oracle RAC Performance Diagnostics and Tuning - page 172

Node ssky1l4p1 has received several alerts compared to many of the other servers in the cluster. All alerts
indicate high I/O wait times, indicating that there is an I/O bottleneck on these devices from this specific node.
Bottlenecks such as these indicate I/O problems on the device itself or the components that connect these servers to
these disks. Since this is one of the few nodes reporting this problem, the I/O problem may not be with the physical
device itself. Further investigation on the I/O paths to the SAN indicated that the system administrators configured
only one HBA on node prddb1 while the other servers had four HBAs configured.
[root@ssky1l4p1 ~]# /sbin/powermt display
Symmetrix logical device count=0
CLARiiON logical device count=0
Hitachi logical device count=0
Invista logical device count=0
HP xp logical device count=21
Ess logical device count=0
HP HSx logical device count=0
==============================================================================
----- Host Bus Adapters --------- ------ I/O Paths ----- ------ Stats ------
### HW Path Summary Total Dead IO/Sec Q-IOs Errors
==============================================================================
4 lpfc optimal 84 0 - 0 0
[root@ssky1l4p4 ~]# /sbin/powermt display
Symmetrix logical device count=0
CLARiiON logical device count=0
Hitachi logical device count=0
Invista logical device count=0
HP xp logical device count=21
Ess logical device count=0
HP HSx logical device count=0
==============================================================================
----- Host Bus Adapters --------- ------ I/O Paths ----- ------ Stats ------
### HW Path Summary Total Dead IO/Sec Q-IOs Errors
==============================================================================
2 lpfc optimal 84 0 - 0 0
3 lpfc optimal 84 0 - 0 0
4 lpfc optimal 84 0 - 0 0
5 lpfc optimal 84 0 - 0 0
HBA is the primary gateway between the node and the physical disks located on the SAN. Due to limitations on
the hardware, it has been observed on several occasions that the number of slots for the host bus adapter (HBA) and
the network interface card (NIC) are insufficient to provide a good I/O capacity. The number of ports on a switch
and the number of controllers in a disk array should also be considered while tuning I/O capacity. Care should be
provided to ensure that the number of HBAs is equal to the number of disk controllers. Using the disk controller slots
to accommodate more disk arrays will have a negative impact on the total throughput.
For example, on a 16-port fiber channel switch, the ideal configuration would be to have 8 HBAs and 8 disk
controllers, giving a total throughput of 8 * 200MB 3 = 1.6 GB/sec. Now, if the number of HBAs is reduced to 4 to
provide room for additional storage, then the total throughput drops down by 50% (4 * 200 MB = 800MB/sec).
3 Assuming the maximum theoretical payload of 2GB/s Fiber Channel is 200MB/sec.