CX4 RG configuration

i believe this is a known issue depending on the level of your FLARE code. the warning should not prevent you from completing the RG creation.

You can create a RAID10 group using disks from that enclosure and another, but If you create a RAID Group with drives from BUS 0 ENCL 0 and any other enclosure you run risk of data corruption in the event of a power loss. This is because the drives in BUS 0 ENCL 0 are powered through the SPS which will keep the enclosure online long enough to shut the array down gracefully in the event of a power failure. So some of the disks will be offline for that RAID Group during a power failure.

Since you are creating a RAID10 group with half the disks in one enclosure, and half in another enclosure, you are minimizing the possibility of data loss in the event of a power outage (whereas if you were using RAID5 you would lose data). But I would think you are still at risk and you would be better off to use drives from another enclosure.

For example, if you take one disk from bus 0 enc 0 and one disk from bus 0 enc 1 (creating a R10), you have one disk in the first DAE - this disk is powered by the SPS. If you lose power, the SPS will keep bus 0 enc 0 powered on until the cache is dumped to disk. But the disk in bus 0 enc 1 has lost power and has shut down. If there was a write in flight to the disk in bus 0 enc 1, you do not have any way to determine if the write completed or not. When this occurs the state of the raid group is now unknown and on power restore, the raid group will need to be re-built - this is for Raid 1/0.

glen

Thanks for the example Glen, based on that I would not even try and use a RAID10 group with half of the disks in bus 0 encl 0. The risks are too great.

Thanks guys for the response.

I get an alert when choosing disks from ex Bus 0 Encl 1 and Bus 1 encl 0 and so on,so I guess I can just ignore the alert for RG creation but make sure our Data center doesn't loose power at any time.

Thanks Glen and Aran for your suggestions on creation of RG 1+0.

so what happens if you create a striped MetaLUN where one of the components reside in RG on B0E0, data loss ?

I would think massive data loss

That would be like a RAID5 group losing more than one drive at a time right?

Uh-oh, i have a component LUN in RG on B0E0 . While we have datacenter UPS and generators ..i am not taking any chances, Lun migrator to the rescue (too bad we can migrate component LUNs only). Now if this is a planned shutdown of CX4, array is still getting datacenter power ..it should be fine ?

I would think it would be fine because in a planned shutdown you switch off power using the SPS power switch which prompts the array to start the shutdown process. This will power down the SPE and B0E0. Even though the remaining enclosures attached to the array are still powered up, they are dumb enclosures at that point and the shutting down of B0E0 will not be breaking a RAID set.

I want to clarify that a "split mirror" config as being discussed earlier in this thread is not recommended, but such a config is _not_ in immediate risk of data loss or data corruption that could arise from a simple power failure.

The scenario we are seeking to avoid is inducing Rebuild operation(s) across 1/2 of the drives in the mirror (RAID 1, or RAID 1/0). Consider the simplest example - a RAID 1 pair with one drive (call it Drive0) in Bus 0 Enclosure 0 and the other drive (call it Drive1) in some other enclosure. If such a config experiences a power failure, Bus 0 Enclosure 0 is going to run for longer than any of the other enclosures. The array sees Drive0 continuing to operate while Drive1 is inaccessible - this can result in the array considering Drive1 to the "faulted", and it will be marked as needing a Rebuild. This is the situation/configuration that we recommend avoiding (as we certainly prefer to maintain redundancy at all times). However, if it does happen... any in-progress IOs will still be completed to Drive0, and the information on Drive0 will be intact; it will not be corrupted or discarded. When the array powers back up, a Rebuild will be performed, and the contents of Drive0 will be rebuilt onto Drive1, restoring the mirror to full redundancy.

So the risk of a "split mirror" config is that a power failure could result in the mirror being broken, requiring a Rebuild operation to restore redundancy. We recommend avoiding that, but such an event, if it occurs, will not corrupt or discard the data contained within the RAID group.

In the more complex/nuanced case of a metaLUN, where some component LUNs are SPS protected and others are not, data integrity is still maintained. Note that the array does not acknowledge writes back to a host server until either; the data is in Write Cache, or the data has been successfully written to disk. If it's a write from Write Cache that gets interrupted by a power failure, that operation will be re-executed when the array powers back up. If it's an un-cached write that gets interrupted, the host will not receive an acknowledgement of the write, and normal host error handling/recovery used for any sort of power failure scenario would occur. The key point is - any write operations that the array acknowledged completion of, are maintained (they are not corrupted or discarded).

Regards,
DGM

P.S. For additional discussion, see the "Binding with DAE0 drives" segment in the CLARiiON Best Practices document.

DGM,

so what would happen in my scenario where i have a 2 component striped MetaLUN where components 0 is in 4+1R5 RG on B0E0 while component 1 is in 4+1R5 RG some other Enclosure ? Also would the outcome be different if this was a concatenated MetaLUN ?

Thank you

dynamox,
It's the same as outlined above - any write operations within Write Cache that don't complete will be completed when the array powers back up, and any write operations that are non-cached that don't complete will be subject to normal host power failure handling. This guidance is applicable independent of what enclosure a LUN or metaLUN component resides in, and independent of whether a metaLUN is striped or concatenated.

Regards,
DGM

Thanks DGM,

so worth case scenario i will have crash consistent data on my MetaLUN but i will not lose data that was already distaged to disk?

dynamox,
You will have a crash consistent image on any LUN or metaLUN. Any writes that were acknowledged by the storage system (writes into Write Cache or uncached writes to disk, which were reported back to the host as being completed) will persist through the power failure.

Regards,
DGM