Original Question:
We are using several T64 v5.1 systems connected with HSG/HSZ controllers in
multi bus failover mode.
When looking at the disks (with hwmgr) you see several valid paths to a
certain disk. As far as I know only one path to a certain disk is used at
any time. When the primary path to a disk is gone, the system will fail over
the traffic to another valid path. The system takes the first path that it
sees as the default valid path. Is there any way to change this in order to
use different default paths to make sure the load is equally divided over
all existing paths. In other words we would like to know if there is a
possibility to change the "First valid path" column in the "hwmgr -show scsi
-id xxx" command to another path.
---------------------------------------------------
Thanks to Jan Mark Holzer from Compaq for his detailed answers:
the operating system will use multiple path to underlying storage if the
storage subsystem does support it. The HSG80/HSZ80 controllers do
support multiple path access to ndividual units , the HSZ50/40 only support
single path active (ie. the addtl. paths are only used for failover).
On the controller level it depends on the controllers capabilities on
how an individual unti can be accessed. On the HSG80/HSZ80
controllers a given unit can only be served by a single controller (ie. one
out of a dual-redundant pair). However from the OS perspective we'd be still
using multiple path to a single unit if the OS does see multiple paths.
Remember that the output of the "hwmgr -show scsi -full" can be somewhat
confusing and miss-leading as the value 'valid' doesn't tell you whether the
pass is active or standby it just tells you the path is available
for the operating system to use .
You can use
"hwmgr -get attr current -cat disk|grep -E
"dev_base_name|path_state|path_xfer"
to see the current state of all available paths and how much I/O has been
sent through an individual path. The state reported can be either 1
(active), 8 (failed), 2 (standby) or 4 (do not use).
"hwmgr -show scsi -full" will give you path states as well.
Based on the system configuration we'll always use a local path over
a "remote" path on GS80/160/320, that's why we highly recommend to have at
least one adapter per QBB/RAD on a GS series system.
After asking if this load balancing was also present in trucluster
configuration he answered:
yep load balancing is independant of TruCluster . In a TruCluster each node
does it's own I/O load balancing. If RAW devices or DirectIO is used than
each node will issue I/O request directly to storage which has a direct path
to the node (for DirectIO there are a few cases where I/O will still use the
CFS server, ie. I/O write operations which are storage allocating writes
(ie. file extents/creates/resize operations).
Regards,
Gert-Jan Hilbrands
------------------------------------------------
Atos Origin Telco Services, Unix support
050-5882781 / 06-51584767
gert-jan.hilbrands_at_atosorigin.com
Received on Fri Nov 30 2001 - 14:17:28 NZDT