Disk Enclosures
1753962 Members
7713 Online
108811 Solutions
New Discussion юеВ

Re: IO_TIME OUT of Hard Disk

 
Vivek Pendse
Occasional Advisor

IO_TIME OUT of Hard Disk

Hi,

I want to know what is the exact use of changing the IO_TIME of the harddisk that we changes by the pvchange command. How it affect the system when we change the IO_TIME of the LUNS?

Vivek
5 REPLIES 5
IT_2007
Honored Contributor

Re: IO_TIME OUT of Hard Disk

IO_TIME is the disk response time for I/O activity. Never change timeout value from default unless you have performance issues on the server. We did this on Oracle servers since performance from DISK was too quick. So we increased to complete IO on the disk.
Vivek Pendse
Occasional Advisor

Re: IO_TIME OUT of Hard Disk

Hi,

But what will be the maximum and minimum value of IO_TIMEOUT of Hard Disk ?
And by increasing the value does it improves the performance ?

Vivek Pendse
IT_2007
Honored Contributor

Re: IO_TIME OUT of Hard Disk

It depends on your application. EMC recommended for our environment to set 90seconds from default value.

-t IO_timeout Set IO_timeout for the physical volume to the
number of seconds indicated. An IO_timeout
value of zero (0) causes the system to use
the default value supplied by the device
driver associated with the physical device.
IO_timeout is used by the device driver to
determine how long to wait for disk
transactions to complete before concluding
that an IO request can not be completed (and
the device is offline or unavailable).
Serviceguard for Linux
Honored Contributor

Re: IO_TIME OUT of Hard Disk

Is this mis-categorized? I don't see anything about IO_TIMEOUT for for LVM on Linux but I do see it for HP-UX.
Florian Heigl (new acc)
Honored Contributor

Re: IO_TIME OUT of Hard Disk

IF we're talking about HP-UX LVM:

(PV_)IO_TIMEOUT is the time the OS allows a disk access to complete, if it doesn't complete within that time, the PV's status in it's volume group will change to unavailable, the IO will fail back to the issueing application and You'll get ugly error messages in dmesg.

Depending on the type of IO You might also lose a bit of data :)

Raising the timeout only makes sense *IF* there is a good chance this will help a potential IO to complete; in reality, if the IO doesn't complete within 2 seconds You already have a BIG issue and changing the default (30s) to a higher value like 90s or 180s (EMC^2 choses to recommend either of those at times) might not always help.

In the EMC^2 context this will be advised if You use Clariion / CX active/passive arrays as there are conditions that might delay an IO for that long. This still is a big issue, but one You payed for :)
yesterday I stood at the edge. Today I'm one step ahead.