Hello All,

I have been looking at the Adaptec 3805 Raid Controller (8 channel). In the specifications it refers to the additional battery backup module (ABM-800). I was wondering why would you require this module, is it just to pretect the cache in the case of a power cut?

Thanks in advance

Karl

You nailed it on the head... If the machine turns off and the controller is still writting to the drive or has something in the cache (typically in the 128mb ram buffer) you can loose data.

I own the 3805 and use it without the backup, however, i use a good UPS which will act the same sort of way as its setup to shut the machine down in the event of a power outage. As windows shuts down the controller will write everything in cache in disk.

A BBU does more than just protect the cache against power failure. If you have an application crash that blue screens, or someone hits the reset button or power switch, you can lose data in the cache even with a UPS attached.

The BBU is specifically to protect the cache integrity under all conditions, not just power failures.

SomeJoe, if power is still going to the card in the event of a BSOD would the cache still not be written as the card is dedicated controller. Has its own CPU and functions seperately from the OS. So data that has been sent to the card will be written?

Well, if the motherboard resets, the BIOS and south bridge chips will send hardware reset signals to all slots, which will cause the RAID card to reset and reinitialize. Depending on the motherboard design, this may also briefly interrupt power to the card. At minimum, all internal state of the RAID controller is lost, and it will go through power-up self-tests and drive initialization.

As long as the battery has kept the cache contents, the RAID controller will identify the dirty cache, and begin flushing it to the array(s) once they're finished with the power-up and initialization sequence.

The key is that a UPS doesn't always guarantee that the card gets a constant supply of power under all conditions. The BBU, because it's specifically for the RAID controller's cache, can guarantee that.

The battery also gives the option of actually moving the RAID card and drives to a different computer should the server physically die. If the motherboard all of a sudden shorts out and crashes, the BBU will actually keep cache contents alive on the card long enough for you to move the card and drives to a different motherboard and power it back up.

Also remember that if the cache data is gone sometimes a delayed write failure can corrupt the entire array not just the small amount of data in cache. So you could lose all the data in the array to corruption. I have seen this happen on a few occasions.

Ok... I've got to ask... Does all of this apply to me if I use a high-end external raid subsystem (infortrend) with a redundant PSU & a good UPS like chookman says?

I was basically thinking the same thing that he (chookman) was: as long as my APC UPS shuts down my PC, then the external raid subsystem will finish writing the data in its cache before it eventually shuts down when the UPS battery runs out.

What is a delayed write failure? The possibility of my entire raid 6 array becoming corrupted really scares me. - I really don't want to spend nearly $200 on a BBU when I already have a UPS, but I will if I have to....

I'm not going to be really long with this explanation, and you can Google for a better understanding. If you have write caching turned on, and most people do use it or the performance will suffer, then things are written to a cache and then written to the actual disk/array later. If the OS is trying to write to something and that something goes away either before (not as bad) or during a write (the really bad one) you can lose data. The delayed write has failed. Now think about if the file system itself gets corrupted during this write failure. You can lose all data in the array. The machine itself can also experience a crash or sudden power off for various reasons.

Joe lists a few reasons above as to why a UPS is not a "best" solution too.

edit: I cant spell