UPS configuration: HP-UX and EMC Clariion

[Jens Ebert]

Perhapes anyone faced the same szenario and has some information for us:

Two HP-UX 11.0 servers (L1000/L3000) are sharing one EMC Clariion disk array. Both have a separate UPS system, the EMC is connected to one of these.

Power failure: First the servers have to shut down their applications and halt, the Clariion has to be available and shut down afterwards.

Power restore: First the Clariion has to be ready before the servers start to boot.

How can this be archived? Up to EMC (they are of no help at all!) there is no solution to have communication take place between the systems.

[Edward Thibeault]

Additional information is required to answer this properly -

1. What type of EMC/Clariion array?
2. Is this a SAN or an FCAL configuration?
3. What are you tring to acomplish with this
configuration?

[Jens Ebert]

We have a FC4500 with two sorage processors (SPA/SPB) installed. Both are directly connected via FC cables to the L-class servers. So both servers share the Clariion, using parts of the disks installed.

All works fine, but right now there is no "power failure szenario" that was available with old EMC Centriplex systems ore HP disk arrays.

[Patrick Wessel]

Jens,
The most elegant way would be to use an UPS that switches the outlets at different times on.
Otherwise you have to include a sleep in the /sbin/lvmrc to give the diskarray time to spin up before the hp-ux expects the disks.

Example for a 20 second sleep:

#vi sleep20.c:

#include
main()
{
sleep(20);
}


compile and copy to /sbin:
# cc -Wl,-a,archive sleep20.c -o sleep20
# mv sleep20 /sbin/sleep20

2) include in /sbin/lvmrc


.
.
.
default_vg_activation()
{
if [ -r /etc/lvmtab ]
echo " Waiting for Disk Array to spin up..."
/sbin/sleep20
then echo "Waited 20 seconds for Disk Array to spin up"

fi
}
.
.
.

[Jens Ebert]

Patrick,
we do not know in detail how long the Clariion will need to recover from a power failure. If we use a constant time to wait, this has to be "long". But in this case the system boot always will be delayed even if there was no power failure.

It should be possible to ask the disk array for it's status and to decide if the boot process has to be delayed or not.

Nobody else has this problem??

Thank you, Jens

[Edward Thibeault]

The Clariion array need 3 minutes to fully power up and make all devices available to the connected system. During a power fail situation the system (Clariion) has its own SPS (standby power supply) so it can shed cache. This opreation will take no more than 90 seconds, then the disk array will shut down.

[Jens Ebert]

Beside the delay of a fixed period of time, is there no command-line utility for the Clariion, that can be used to check it status? If this status would report "ready, up to use", the HP-UX servers could continue to boot.

Is this kind of utility available?

[Jens Ebert]

On HP-UX I found the command-line utility "navicli". This provides some "getxxx" commands like getcrus, getcontrol, getconfig, getagent, getcache, ...

Does someone know if I can use the output of one of these commands upon boot to find out if the disk array is fully operational to that the lvm can access the disks?

[harry d brown jr · Accepted solution]

Jens,

You should REARRANGE your POWER feeds!

ONE L6-30P from your Clariion into one UPS#1, and the other L6-30P into UPS#2. You'll probably need to change power cords.

Then do the same for your L-class servers - which will RUN on a SINGLE power supply! The L's are designed to have three power supplies and should be wired that way to UPS's. Have one PDU into UPS#1 and another PDU into UPS#2.

As for booting after a power failure, I always do MANUAL intervention on systems that rely on EXTERNAL storage unless I can be guaranteed that the storage device will always be ready. I would not want to screw around with having to reboot the HP's again if they timeout because the Clariion (or any disk subsystem for that matter) isn't ready.

The Clariions should be up and running WAY before the HP's are ready to boot, so you shouldn't have any issues!


live free or die
harry