Immediate_Report default_disk_ir

[Ashwin_4]

Hi,
default_disk_ir is set to 0 (Zero) in kernel. And the output of scsictl shows,
# scsictl -m ir -m queue_depth /dev/rdsk/c4t1d2
immediate_report = 1; queue_depth = 8

Whether I should set default_disk_ir to 1 to enable the write cache.
Thanks

[Stefan Farrelly]

There is a risk versus performance issue here. With HP-UX 10.X and 11.0 the default is default_disk_ir = 0, the reason being if it was set to 1 (to increase performance) you increased the risk of a filesystem corruption in the event of a system crash.

In 11.11 and on default_disk_ir = 1 by default, which means the problem of a possible corruption in the event of a crash has been fixed, so its on by default to enhance performance.

If youre running 11.11 its on already, if youre running 10.X or 11.0 I would leave it set to 0 as its not worth the risk of a possible corruption just to get a little extra performance.

[Kent Ostby]

In addition to the risk of corruption for a disk failure, the use of immeadiate reporting early on occassionally caused system panics of the type:

SCSI: Unrecovered Deferred Error

The reason for this is that the system assummed the I/O was successful (because of IR) and then it would fail.

When the deferred error was picked up on a later I/O to the same device, the system would panic since the kernel would then (rightfully) decide that it was in an inconsistent state.

If IR was turned off, then the system would retry the I/O at the software level and would hopefully not get into an inconsistent state.

Best regards,

Kent M. Ostby

[Dietmar Konermann]

Just some additions...

By default the default_disk_ir tunable is still 0, even with 11.23:

# kctune -q default_disk_ir
Tunable Value Expression
default_disk_ir 0 Default

If I remember correctly, then default_disk_ir gets set to 1 on workstations.

The "Panic: SCSI Unrecovered Deferred Error" has been removed long time ago... beginning with 11.11 and e.g. for 11.00 with PHKL_20208.

The current bahaviour is to simply block all I/O to the device that has reported the deferred error until it get's closed/re-opened.

Best regards...
Dietmar.

[Stefan Farrelly]

Dietmar,

On ALL our 11.11 installations, even our new servers and Superdomes (installed by HP) all have default_disk_ir = 1

Ive just installed new 11i server with the Sep2003 cd and its also set to 1.

[Bill Hassell]

Typically, default_disk_ir=1 should only be set when *every* disk connected to the computer has a battery backup. The reason is that the cache on the disk may contain critical information that has not yet been written. A powerfail will lose that data and potentially corrupt filesystems stored on the disk. default_disk_ir only affects simple disks (JBODs) since disk arrays have buffering internally that is either always on or can be setup to be on at all times. And most disk arrays have a battery backup built-in that lasts long enough to write all data in the buffer to the appropriate disks.

[Dietmar Konermann]

Stefan,

from the kernel perspective the default is 0 on 11.11 (active when not explicitely set):

# grep -e PHKL -e default_disk_ir /usr/conf/master.d/core-hp>
* @(#) core-hpux $Date: 2002/07/17 05:55:44 $Revision: r11.11/1 PATCH_11.11 (PHKL_27431)
default_disk_ir DEFAULT_DISK_IR 0

So I assume the parameter gets set in your /stand/system files somhow. Maybe some Ignite setting? No idea, to be honest. Just checked on several systems here in Ratingen... didn't find the paramter in any system file, -> 0.

BTW, the new kernel man pages of 11.23 explain tunables pretty detailed.

http://docs.hp.com/cgi-bin/onlinedocs.py?mpn=B2355-60103&service=hpux&path=00/40/4038&title=HP-UX%20Reference%20%2811i%20v2%29

Best regards...
Dietmar.

[Stefan Farrelly]

Youre right Dietmar,

once our servers are built with latest patch bundles/diags/apps it includes the EVA/Securepath software which sets it to 1 - which matches Bill's comments about disk subsystems with backup means it can be set to 1 safely.