RAID using write-back (lazy writes)

[cbh]

"You've configured your RAID array to use write-back? Man, if someone in the same street as you so much as sneezes, you'll lose your filesystems, your crops will fail and your firstborn will die."

Okay, that might be a little melodramatic, but that's my understanding of the so-called evils of using write-caching with hardware RAID (my definition of "hardware raid" being anything where the operating system doesn't have intimite knowledge of what it's up to at any particular instant). I'm curious as to what thoughts and experiences people here have of that type of issue; this follows on from a comment I made in someone else's thread that my Adaptec 2410SA's RAID-10 performance is not exactly earth-shattering, with sequential writes in the 7MB/s ballpark: I like the idea of secure data, but a bit of performance wouldn't go amiss either.

So I guess my questions are:
- is write-back going to kill my data if the cable falls out the back of my machine?
- is battery-backed write-back any better? If it's as safe as write-through, then given the performance difference would I be better off replacing the 2410SA with an LSI Logic or 3Ware jobbie?
- since I'm not short of CPU power, should I forget "hardware" raid altogether and wait for the bugs in FreeBSD's gvinum (and Linux's and Windows XP's equivalents on their respective partitions) to be ironed out?

 

[UICompE02]

- is write-back going to kill my data if the cable falls out the back of my machine?

Well, it is very likely that your data will become corrupted or lost if it hasn't been written to the disk and the connection is lost.

- is battery-backed write-back any better? If it's as safe as write-through, then given the performance difference would I be better off replacing the 2410SA with an LSI Logic or 3Ware jobbie?

This is precisely the reason battery-backed RAID adapters were created. If the power fails or something else happens, the data is maintained in the RAM until the controller gets a chance to commit it to the disk.

 

[cbh]

Well, it is very likely that your data will become corrupted or lost if it hasn't been written to the disk and the connection is lost.

In particular it's the metadata that I'm worried about: a file or two I can probably handle (although that would be annoying) but larger-scale filesystem corruption is another matter entirely, certainly not something I want to experience again (last time it happened was using async-mounted systems on Linux: I had to learn the hard way that it was a Bad Idea!)

This is precisely the reason battery-backed RAID adapters were created. If the power fails or something else happens, the data is maintained in the RAM until the controller gets a chance to commit it to the disk.

That sounds positive. Would you say that the battery-backed writeback should offer as much data integrity as writethrough? (The card itself being a potential point of failure notwithstanding)

Another question (sorry, I seem to have lots of them regarding this topic!): does the card's write cache still give a significant performance boost even with the drive's writeback cache turned off? I'm very unwilling to use writeback on the drives themselves.

 

[Great_Melinko]

i <3 my netraid card with battery backup