Unanswered: IO stalls to 100% busy when accessing a SAN disk with backup of Informix dbspace
Having tried many things I keep having a problem with a Emulex 10000 HBA. When doing a backup of our database with Netbackup, we see backup is a lot slower at 1 HBA.
The symptoms are that IO's go OK until a certain point where IO's 'stall'. With iostat one sees the device in question is giving 100% busy while IO's as well as service times remain 0. This lasts for 30-50 seconds and then IO's are picking up again. See listing below. When forcing IO traffic to go through the other HBA there are no such busy %.
Anyone could give me a clue what's happening?
Listing of iostat -zxn 5, where the c6 controller is the one having the problem:
The 100% busy remains there for 30-60 secs, then there is a burst of normal IO traffic with 30 MBs/sec for 20 secs, then 100% busy again, etc.
Our configuration is:
- Solaris 9, latest patches applied, on Fujitsu 1500 hardware.
- 2 Emulex 10000 Light pulse cards with latest driver software 6.02h
- Veritas Volume Manager 4.1 with latest service pack MP1
- Veritas Netbackup 5.1 with latest maintenance pack MP4S01
- Datbase: Informix 9.40 FC5XG. Backup goes through Netbackup/onbar scripts. Database is held on raw devices in the San, that is accessed through Veritas Volume Manager.
If anyone can give me a clue, it will be greatly apreciated.