Tuesday, August 16, 2005

iSeries i5 Woes

It is my understanding from speaking with an IBM hardware guy that the new iSeries i5 models (not certain of exact model #'s - could be all of them) are having some fairly consistent issues with the DASD Controller cards related to the cache. Seems that the cards are just dropping out and wiping out the cache. This is not good.

The "fix" is that now you will need to have a mirrored cache card. This will take up a full PCI slot and you will need one of these for every DASD Controller IOP. I also understand that new systems will default with these installed on the order form. Rumor has it that a large bank has lost a system because of this...

3 comments:

Chris Whisonant said...

Found out some more info about this. It's actually referred to as "Auxiliary Write Cache" available May 27, 2005. Here's the IBM spin

"These new features provide a capability between RAID-5 and controller level Mirroring. While RAID-5 protects from single disk failures, various failures to the disk controller card could bring the system down. What's worse, depending on the type of failure, the card may need to be replaced, and any data on the cache on that card may not be available. This can potentially cause a situation where the system needs to be reloaded (quite time consuming). With controller level mirroring, the failure of a disk controller card is less eventful. Since the card is mirrored, the system keeps on running (potentially with less performance). The failing card can be replaced via concurrent maintenance and the system will re-synchronize the data on the disks.

With these new features, if a controller card fails, the system will still be unusable until the controller card is repaired/replaced. However, a copy of the cache that was contained on the card being replaced is still available on the second card. When the failing card is replaced, the cache is reloaded from the second card, and the system is restarted. Information contained in the card is written to disk, and the system reload will usually be unnecessary. This is still not as good as controller mirroring (where the system remains running), but is usually much less expensive. You and your customers need to discuss the potential cost of an outage that could last several days. An informed decision can then be made."

Chris Whisonant said...

David, hopefully you'll be OK. From my understanding they default to ordering the Aux Write Cache. Check to see if you have either the 5580 or 5581 Feature Codes on the order.

"These features are a combination of two PCI cards (taking two PCI slots). The #5580 feature includes a #2780 and an additional card (and cable). The #5581 feature includes a #2757 and an additional card (and cable)."

Chris Whisonant said...

Oh, thanks for the link. Hopefully you'll enjoy your 520s! We're still on 820s here, but hopefully we'll be able to get to i5 next year.