Disk Enclosures
1748354 Members
4910 Online
108763 Solutions
New Discussion юеВ

IO Timeout for Fibre Channel Attached Disks

 
SOLVED
Go to solution
Justin Willoughby
Regular Advisor

IO Timeout for Fibre Channel Attached Disks


I notice the timeout for switching between fibre channel disks (alternate links) is about a minute. I know this can be changed with pvchange but it says the default is to use the driver's timeout. I have two A6685A cards.

Are there any cons to setting the timeout lower then the default? A minute seems like a lot of time to fail to the next disk/path.

Also, as the disks are on a SAN (IBM Shark) and not locally attached should I disable bad block relocation? I assume so but wanted to double check. If I should disable this, how do I do it?

Thanks!

- Justin
5 REPLIES 5
Tom Geudens
Honored Contributor
Solution

Re: IO Timeout for Fibre Channel Attached Disks

Hi Justin,
All your questions are HARDWAREdependent, so check with your vendor (IBM in this case). In my environment (EMC ClarIIon and Symmetrix) we use the following :
# Timeout parameter
pvchange -t 180 /dev/dsk/
# Bad block relocation
lvcreate -L -r N -n lv_name vgname

But like I said, it depends !
Regards,
Tom
A life ? Cool ! Where can I download one of those from ?
Bernhard Mueller
Honored Contributor

Re: IO Timeout for Fibre Channel Attached Disks

Justin,

don't know about the shark but typically with EMC or RAID boxes in general you should disable Bad Block relocation since the Array has its own way of dealing with such things.

Also timeout ist usually set to a higher value than default....

options to use:

pvchange -t 180 /dev.....

lvcreate -r N .....

Regards
Bernhard
Ashwani Kashyap
Honored Contributor

Re: IO Timeout for Fibre Channel Attached Disks

Its a good idea to disable bad block relocation since the array takes care of it automatically and you don't need another software system .

Generally I have seen a timeout of 180 sec is very good , with disks on fibre channel . I haven't dealt with IBM SHARK , but on EMC , hitachi , XP256 and Compaq SAN , the value of 180 sec , is good enough .
Justin Willoughby
Regular Advisor

Re: IO Timeout for Fibre Channel Attached Disks


Anyone know why the timeout is so long? Failing between alternate links would cause the application to possibly freeze for a minute or two while this happens. I assume there is a reason for the timeout to be as high as it is (as the default is or as suggested here).

Thanks,

- Justin
Michael Tully
Honored Contributor

Re: IO Timeout for Fibre Channel Attached Disks

We tested this very issue just before going live on an EMC 8530, and the failover to the alternate disks was around ten seconds. I can't speak for the IBM shark. We also use 180 second timeouts and 'N' bad block relocation. Suggest you contact IBM to find out their recommendation. I would be surprised if it is much different to the other manufacturers. Lowering the timeout from the default could cause you more headaches.....
The timeout is set high, so the disk array can complete it's task internally without having to report errors like "SCSI lbolt" back into your system log.
Anyone for a Mutiny ?