sd naming issue with SLES 10/Qlogic HBA

English Hardware
#1

Hi folks,

I’ve been working with some SCSI devices and confused about the sd* naming in SLES 10.

I found on some systems, the sd* names are assigned sequentially from on SCSI host to another, while on some other systems, the names look random as if they are assigned in parallel.

===================================================
Parallel assigned sd* names

[3:0:0:0] disk HITACHI DF600F 0000 /dev/sdb
[3:0:0:1] disk HITACHI DF600F 0000 /dev/sdc
[3:0:0:2] disk HITACHI DF600F 0000 /dev/sdd
[3:0:0:3] disk HITACHI DF600F 0000 /dev/sde
[3:0:0:4] disk HITACHI DF600F 0000 /dev/sdf
[3:0:0:5] disk HITACHI DF600F 0000 /dev/sdh
[3:0:0:6] disk HITACHI DF600F 0000 /dev/sdp
[3:0:0:7] disk HITACHI DF600F 0000 /dev/sdr
[3:0:0:8] disk HITACHI DF600F 0000 /dev/sdt
[3:0:0:9] disk HITACHI DF600F 0000 /dev/sdy
[3:0:0:10] disk HITACHI DF600F 0000 /dev/sdal
…
[3:0:0:100] disk HITACHI DF600F 0000 /dev/sdod
[3:0:0:101] disk HITACHI DF600F 0000 /dev/sdof
[4:0:0:0] disk HITACHI DF600F 0000 /dev/sdg
[4:0:0:1] disk HITACHI DF600F 0000 /dev/sdi
[4:0:0:2] disk HITACHI DF600F 0000 /dev/sdj
…

===================================================
Sequentially assigned sd* names

[5:0:0:1] disk HP HSV210 6000 /dev/sdc
[5:0:0:2] disk HP HSV210 6000 /dev/sdd
[5:0:0:3] disk HP HSV210 6000 /dev/sde
[5:0:0:4] disk HP HSV210 6000 /dev/sdf
[5:0:0:5] disk HP HSV210 6000 /dev/sdg
[5:0:0:6] disk HP HSV210 6000 /dev/sdh
…
[5:0:0:90] disk HP HSV210 6000 /dev/sdcn
[5:0:0:91] disk HP HSV210 6000 /dev/sdco
[6:0:0:0] storage HP HSV210 6000 -
[6:0:0:1] disk HP HSV210 6000 /dev/sdcp
[6:0:0:2] disk HP HSV210 6000 /dev/sdcq
…
[6:0:0:90] disk HP HSV210 6000 /dev/sdga
[6:0:0:91] disk HP HSV210 6000 /dev/sdgb
[7:0:0:0] storage HP HSV210 6000 -
[7:0:0:1] disk HP HSV210 6000 /dev/sdgc
[7:0:0:2] disk HP HSV210 6000 /dev/sdgd
…

If the HBAs are of the same model and driver is also the same, what a explanation could be to this difference?

A friend said in SLES 10, the SCSI device scan is initialed by the HBA driver and qla2x00_rescan_fcports does the scan in threads for each SCSI host in parallel.

If that’s the case, how can you explain that on some system, the names are sequentially assigned?

Thanks a bunch!

Best,
Kenny

#2

To cut to the chase, with SCSI devices you should not depend on the device pathname, as spin up delays etc will affect the order of detection. Instead, mount by label.

#3

Thanks friend.

I know that certain utilities such as udev can avoid the naming issue.

But the situation is that we have some old software that is sensitive to the sd* names.

I’m just wondering why on some hosts that the names are sequential.

Is it about the driver or also array related?

Thanks a lot.

Kenny

#4

Just differences in timing. No guarantees.